SlideShare a Scribd company logo
1 of 41
Download to read offline
Debugging AI
Dr. Christian Betz
3Photo by Gratisography from Pexels
„Debugging AI“ is deliberately
ambiguous.
What drives the current AI hype?
4
5
Compute Power
https://www.youtube.com/watch?v=0ibVhtuQkZA
6
More data
Photo by Negative Space from Pexels
7
New business opportunities

(plus attention)
By Dllu - Own work, CC BY-SA 4.0, https://commons.wikimedia.org/w/index.php?curid=63450446
8
and language processing.
28 Chapter 2 – Why is AI important?
AI is important because, for the first time, traditionally human
capabilities can be undertaken in software inexpensively
and at scale. AI can be applied to every sector to enable
new possibilities and efficiencies.
40 Chapter 3 – Why has AI come of age?
Specialised hardware, availability of training data, new
algorithms and increased investment, among other factors,
have enabled an inflection point in AI capability. After seven
false dawns since the 1950s, AI technology has come of age.
constraints of human experience
82 Chapter 6 – The war
While demand for AI professiona
winners and losers are emerging
Part 3: The AI Disrup
96 Chapter 7 – Europe’s
The landscape for entrepreneurs
AI startups are maturing, bringing
industries, and navigating new op
While the UK is the powerhouse o
France may extend their influence
https://www.mmcventures.com/wp-content/uploads/2019/02/The-State-of-AI-2019-Divergence.pdf
Speed of development
https://www.mmcventures.com/wp-content/uploads/2019/02/The-State-of-AI-2019-Divergence.pdf 9
years (Fig. 23), to an estimated $15bn in 2018 (CB Insights,
MMC Ventures).
Today’s leading technology companies – including Apple,
Amazon, Facebook, Google, IBM, Microsoft and Salesforce
– are also spending heavily on research and personnel to
develop and deploy AI. Internal corporate investment on AI,
among just the top 35 high tech and advanced manufacturing
companies investing in AI, may be 2.0x to 4.5x greater than the
capital invested by venture capital firms, private equity firms
and other sources of external funding combined (McKinsey),
further catalysing progress.
have increased fifteen-fold
in five years, to an estimated
$15bn in 2018.
(CB Insights, MMC Ventures)
Source: CB Insights, MMC Ventures
Fig 23. Venture capital investment in AI has increased 15-fold in five years
0
200
400
600
800
1000
1200
0
2
4
6
8
10
12
14
16
20132012 2014 2015 2016 2017 2018E
Fig. X: Venture capital investment in AI has increased 15-fold in five years
AIdeals
Disclosed Funding (right axis)
Number (left axis)
AIdealinvestment($billion)
High valuation
10
Fear
Photo by samer daboul from Pexels
It’s important to understand…
11
that these factors impact your AI project.
What is AI?
Strong AI vs. Weak AI
AI is used as a generic term for a set of tools to cope with a certain set of
problems.
Machine Learning is a subset of this AI-toolset: „Programming by example“.
Other subsets are knowledge representation, planning, reasoning.
AI uses probabilistic logic instead of boolean logic.
12
13
„The brown quick fox jumps over the lazy dog“
https://www.mcohen.io/2017/machine-learning-explained-in-three-easy-steps/
brown quick
Properties of AI problems
Hard to code „by hand“
• Requires non-formalized knowledge (experiential knowledge)
• Or even not yet existing knowledge
Afflicted with uncertainty (or missing information)
Changes rapidly (making it unreasonable to adopt software
manually)
14
Playing Chess, playing Go
15
f(s)
f(s)
f(s)
f(s)
f(s)
f(s)
f(s)
f(s)
1
0
0
1
Another typical problem: Computer vision
16
This is typically solved using „Deep neural
networks“
Neural networks are simple (sort of):
17
⨁
Input 1
Input 2
Input 3
Output
Weight 1
Weight 2
Weight 3
Deep Networks combine these simple
elements.
18
https://courses.cs.ut.ee/MTAT.03.291/2015_spring/uploads/Main/Artificial%20neural%20network%20for%20image%20classification.pdf
complex patterns in the data as it is propagated throught the network. For example, if the network is
given a task to recognize a face, the first hidden layer might act as a line detector, the second hidden
takes these lines as input and puts them together to form a nose, the third hidden layer takes the
nose and matches it with an eye and so on, until finally the whole face is constructed. This hierachy
enables the network to eventually recognize very complex objects.3
As stated before, the network is able to accurately approximate an arbitrary function by altering its
weights in a systematic way. Initially, the weights are given random values and the network must
be trained in order to find the weigth parameters that produce the desired effect. In order to achieve
Figure 2. An example of neural network layers as feature detectors.4
bles the network to eventually recognize very complex objects.
19Photo by Gratisography from Pexels
digging deeper
into „debugging“
Recap: Properties of AI problems
Hard to code „by hand“
• Requires non-formalized knowledge (experiential knowledge)
• Or even not yet existing knowledge
Afflicted with uncertainty (or missing information)
Changes rapidly (making it unreasonable to adopt software
manually)
20
21
These properties make verification
hard „by design“
Recap: Properties of AI problems
Hard to code „by hand“
• Requires non-formalized knowledge (experiential knowledge)
• Or even not yet existing knowledge
22Photo by Pixabay from Pexels
Problems with unknown truth
For example: Medical classification problem
What is the correct diagnosis?
What is the correct therapy?
You won’t know (maybe until your patient either recovers or dies?)
Same is true for customer support systems. Is you customer satisfied?
You’ll probably only know by loosing him/her as a customer.
23https://towardsdatascience.com/gender-bias-word-embeddings-76d9806a0e17, Photo by rawpixel.com from Pexels
ML replicates bias in the data
Example: Conceptual similarities from word embeddings
With words typically collocated, you can ask you model for
conceptual similarities:
king - man + woman ⇾ queen
Depending on your input corpus, your model will give you
doctor - man + woman ⇾ nurse
Do you really run a sexist, racist chatbot on your website?
AI systems break fundamental patterns we
developed as an industry
• No (or very little) isolation. You need to verify and retrain the
whole system.
• Higher dimension of failure space
• Time intense training cycles (instead immediate feedback cycles)
• Non-stationary nature of ML systems
24http://ai.stanford.edu/~zayd/why-is-machine-learning-hard.html
25https://www.youtube.com/watch?v=piYnd_wYlT8
AI is prone to hacking. So know your tools!
We need …
26
… new quality management approaches
Understand model
requirements
Build the right thing...
27
Model requirements
Check whether the product / project requirements can be met with
the model choices.
Example:

Classify documents
Distinguish Cards (ID, Member Cards), structured documents or
free text? All of them? Or even mixed?
Are you also trying to extract information from these documents?
28
Quality management
on input data
Machine learning is
programming by example.
Make sure your examples
are legit.
29
Use BI, data visualization, statistics, Excel
Implement quality management for your input data.
Visualize, use statistical metrics, use data science on the input data, e.g. using
Tableau, Looker, Excel, whatever suits you)
E.g., to identify bias in the input data, check whether data is balanced, etc.
Use version control on input data (and processing steps) (see dvc.org, that's
not perfect, but what is?). Make sure data is versioned alongside code and
model.
If you create the data from another technical system, manage quality of the
generating code.
Try to document / check for documentation: Which distribution, what features.
30
With supervised learning, you need
supervisors.
Establish panel of judges both on input data labelling and on
outcomes. Due to the non-stationary nature this is is an ongoing
task, not a closed project.
Crowdsource your data labelling or let clickworkers do their job.
QM labelling, again using data visualization, fraud detection (outlier
detection, etc.)
31
Make sure you know the data
Use generated test data with known patterns, because otherwise
you won’t know if you miss whole categories.
32
Cope with
probabilistic logic
The world is not black and
white, it's full of color.
Make sure you see it!
33
Work with test sets
To test for non-binary outcomes (i.e., results with confidence level),
you need to handle test set as opposed to sets of single test
outcomes: For example test for outcome confidence distribution.
For non-stationary systems: establish test monitoring to accept
regression. „Accept new model if result is at least 95% of last
model.“
Do not only test outcomes, but implement inspection tools. E.g., in
RoboCup Simulation map agent movement paths.
34
Work with specific
tools
These depend on your
model requirements
35
These techniques are not generally applicable
Alter real data to fabricate test data with known properties.
Modify input variables and observe model behaviour. Usually small
changes should have small effects.
Sometimes, you even know the direction, amount or characteristic
of effects by applying common sense.
Measure at the right granularity (so the system won't utilize
loopholes)
36
Add a safety net
AI is a tool like any other. It
is created by humans, and
we all make mistakes.
37
Add a safety net
Implement multi-layer security fallbacks for subsets of your
problem (also to be used while testing), like „emergency break
systems“. Test these. Use for testing: If you need your security
fallback too often, your model may be bad.
Implement rules (i.e., code) for parts of your problem.
Just in research: Add explainability, local properties by black-box
tests on the model to verify the „anchor rules" (https://
homes.cs.washington.edu/~marcotcr/aaai18.pdf)
38
© data42 GmbH
Speed / cost benefits of Machine
Learning often only apply if you accept
a non-zero failure rate
Think twice before implementing high stakes AI applications
(E.g., do NOT use AI to screen applications in HR)
39
26.02.18
© data42 GmbH
Derive new knowledge from AI
… to solve new problems.
40
26.02.18
Thank you
- and be curious!
Dr. Christian Betz, data42

More Related Content

What's hot

Dependency inversion using ports and adapters
Dependency inversion using ports and adaptersDependency inversion using ports and adapters
Dependency inversion using ports and adaptersMahfuzul Haque
 
Quality engineering in a world with AI and IoT
Quality engineering in a world with AI and IoTQuality engineering in a world with AI and IoT
Quality engineering in a world with AI and IoTSTePINForum
 
Scaling AI in production using PyTorch
Scaling AI in production using PyTorchScaling AI in production using PyTorch
Scaling AI in production using PyTorchgeetachauhan
 
ACM Chicago March 2019 meeting: Software Engineering and AI - Prof. Tao Xie, ...
ACM Chicago March 2019 meeting: Software Engineering and AI - Prof. Tao Xie, ...ACM Chicago March 2019 meeting: Software Engineering and AI - Prof. Tao Xie, ...
ACM Chicago March 2019 meeting: Software Engineering and AI - Prof. Tao Xie, ...ACM Chicago
 
Reflex and model based agents
Reflex and model based agentsReflex and model based agents
Reflex and model based agentsMegha Sharma
 
A Model-Based Systems Engineering Approach to Portfolio Management
A Model-Based Systems Engineering Approach to Portfolio ManagementA Model-Based Systems Engineering Approach to Portfolio Management
A Model-Based Systems Engineering Approach to Portfolio ManagementElizabeth Steiner
 
EclipseCon France 2015 - Science Track
EclipseCon France 2015 - Science TrackEclipseCon France 2015 - Science Track
EclipseCon France 2015 - Science TrackBoris Adryan
 
Information Technology Research Topics for Postgraduate
Information Technology Research Topics for PostgraduateInformation Technology Research Topics for Postgraduate
Information Technology Research Topics for PostgraduatePhdtopiccom
 
Getting Productive my Journey with Grakn and Graql
Getting Productive my Journey with Grakn and GraqlGetting Productive my Journey with Grakn and Graql
Getting Productive my Journey with Grakn and GraqlVaticle
 
Future is private intel dev fest
Future is private   intel dev festFuture is private   intel dev fest
Future is private intel dev festgeetachauhan
 
Building AI with Security and Privacy in mind
Building AI with Security and Privacy in mindBuilding AI with Security and Privacy in mind
Building AI with Security and Privacy in mindgeetachauhan
 
Decentralized AI Draper
Decentralized AI   DraperDecentralized AI   Draper
Decentralized AI Drapergeetachauhan
 
MLSEV Virtual. Optimization of Passengers Waiting Time in Elevators
MLSEV Virtual. Optimization of Passengers Waiting Time in ElevatorsMLSEV Virtual. Optimization of Passengers Waiting Time in Elevators
MLSEV Virtual. Optimization of Passengers Waiting Time in ElevatorsBigML, Inc
 

What's hot (19)

Demystifying Data Science
Demystifying Data ScienceDemystifying Data Science
Demystifying Data Science
 
What's New in Innoslate 4.3
What's New in Innoslate 4.3What's New in Innoslate 4.3
What's New in Innoslate 4.3
 
Dependency inversion using ports and adapters
Dependency inversion using ports and adaptersDependency inversion using ports and adapters
Dependency inversion using ports and adapters
 
Quality engineering in a world with AI and IoT
Quality engineering in a world with AI and IoTQuality engineering in a world with AI and IoT
Quality engineering in a world with AI and IoT
 
Scaling AI in production using PyTorch
Scaling AI in production using PyTorchScaling AI in production using PyTorch
Scaling AI in production using PyTorch
 
ACM Chicago March 2019 meeting: Software Engineering and AI - Prof. Tao Xie, ...
ACM Chicago March 2019 meeting: Software Engineering and AI - Prof. Tao Xie, ...ACM Chicago March 2019 meeting: Software Engineering and AI - Prof. Tao Xie, ...
ACM Chicago March 2019 meeting: Software Engineering and AI - Prof. Tao Xie, ...
 
Reflex and model based agents
Reflex and model based agentsReflex and model based agents
Reflex and model based agents
 
A Model-Based Systems Engineering Approach to Portfolio Management
A Model-Based Systems Engineering Approach to Portfolio ManagementA Model-Based Systems Engineering Approach to Portfolio Management
A Model-Based Systems Engineering Approach to Portfolio Management
 
H2020-AHTOOLS Use Case 3 Functional Design
H2020-AHTOOLS Use Case 3 Functional DesignH2020-AHTOOLS Use Case 3 Functional Design
H2020-AHTOOLS Use Case 3 Functional Design
 
EclipseCon France 2015 - Science Track
EclipseCon France 2015 - Science TrackEclipseCon France 2015 - Science Track
EclipseCon France 2015 - Science Track
 
Information Technology Research Topics for Postgraduate
Information Technology Research Topics for PostgraduateInformation Technology Research Topics for Postgraduate
Information Technology Research Topics for Postgraduate
 
Innoslate for Academia
Innoslate for AcademiaInnoslate for Academia
Innoslate for Academia
 
Getting Productive my Journey with Grakn and Graql
Getting Productive my Journey with Grakn and GraqlGetting Productive my Journey with Grakn and Graql
Getting Productive my Journey with Grakn and Graql
 
Software bug prediction
Software bug prediction Software bug prediction
Software bug prediction
 
Future is private intel dev fest
Future is private   intel dev festFuture is private   intel dev fest
Future is private intel dev fest
 
Building AI with Security and Privacy in mind
Building AI with Security and Privacy in mindBuilding AI with Security and Privacy in mind
Building AI with Security and Privacy in mind
 
Developing Digital Twins
Developing Digital TwinsDeveloping Digital Twins
Developing Digital Twins
 
Decentralized AI Draper
Decentralized AI   DraperDecentralized AI   Draper
Decentralized AI Draper
 
MLSEV Virtual. Optimization of Passengers Waiting Time in Elevators
MLSEV Virtual. Optimization of Passengers Waiting Time in ElevatorsMLSEV Virtual. Optimization of Passengers Waiting Time in Elevators
MLSEV Virtual. Optimization of Passengers Waiting Time in Elevators
 

Similar to Debugging AI

The State of Artificial Intelligence in 2018: A Good Old Fashioned Report
The State of Artificial Intelligence in 2018: A Good Old Fashioned ReportThe State of Artificial Intelligence in 2018: A Good Old Fashioned Report
The State of Artificial Intelligence in 2018: A Good Old Fashioned ReportNathan Benaich
 
Big Data & Machine Learning - TDC2013 São Paulo - 12/0713
Big Data & Machine Learning - TDC2013 São Paulo - 12/0713Big Data & Machine Learning - TDC2013 São Paulo - 12/0713
Big Data & Machine Learning - TDC2013 São Paulo - 12/0713Mathieu DESPRIEE
 
Big Data & Machine Learning - TDC2013 Sao Paulo
Big Data & Machine Learning - TDC2013 Sao PauloBig Data & Machine Learning - TDC2013 Sao Paulo
Big Data & Machine Learning - TDC2013 Sao PauloOCTO Technology
 
Human-Centered AI: Scalable, Interactive Tools for Interpretation and Attribu...
Human-Centered AI: Scalable, Interactive Tools for Interpretation and Attribu...Human-Centered AI: Scalable, Interactive Tools for Interpretation and Attribu...
Human-Centered AI: Scalable, Interactive Tools for Interpretation and Attribu...polochau
 
UNCOVERING FAKE NEWS BY MEANS OF SOCIAL NETWORK ANALYSIS
UNCOVERING FAKE NEWS BY MEANS OF SOCIAL NETWORK ANALYSISUNCOVERING FAKE NEWS BY MEANS OF SOCIAL NETWORK ANALYSIS
UNCOVERING FAKE NEWS BY MEANS OF SOCIAL NETWORK ANALYSISpijans
 
UNCOVERING FAKE NEWS BY MEANS OF SOCIAL NETWORK ANALYSIS
UNCOVERING FAKE NEWS BY MEANS OF SOCIAL NETWORK ANALYSISUNCOVERING FAKE NEWS BY MEANS OF SOCIAL NETWORK ANALYSIS
UNCOVERING FAKE NEWS BY MEANS OF SOCIAL NETWORK ANALYSISpijans
 
AI Orange Belt - Session 2
AI Orange Belt - Session 2AI Orange Belt - Session 2
AI Orange Belt - Session 2AI Black Belt
 
Explainable AI in Industry (FAT* 2020 Tutorial)
Explainable AI in Industry (FAT* 2020 Tutorial)Explainable AI in Industry (FAT* 2020 Tutorial)
Explainable AI in Industry (FAT* 2020 Tutorial)Krishnaram Kenthapadi
 
Applied AI Workshop - Presentation - Connect Day GDL
Applied AI Workshop - Presentation - Connect Day GDLApplied AI Workshop - Presentation - Connect Day GDL
Applied AI Workshop - Presentation - Connect Day GDLMarc Teunis
 
Machine learning for sensor Data Analytics
Machine learning for sensor Data AnalyticsMachine learning for sensor Data Analytics
Machine learning for sensor Data AnalyticsMATLABISRAEL
 
Model governance in the age of data science & AI
Model governance in the age of data science & AIModel governance in the age of data science & AI
Model governance in the age of data science & AIQuantUniversity
 
Quant university MRM and machine learning
Quant university MRM and machine learningQuant university MRM and machine learning
Quant university MRM and machine learningQuantUniversity
 
Human in the loop: Bayesian Rules Enabling Explainable AI
Human in the loop: Bayesian Rules Enabling Explainable AIHuman in the loop: Bayesian Rules Enabling Explainable AI
Human in the loop: Bayesian Rules Enabling Explainable AIPramit Choudhary
 
Deep Learning - Hype, Reality and Applications in Manufacturing
Deep Learning - Hype, Reality and Applications in ManufacturingDeep Learning - Hype, Reality and Applications in Manufacturing
Deep Learning - Hype, Reality and Applications in ManufacturingAdam Cook
 
FROM BI TO APPLIED AI
FROM BI TO APPLIED AIFROM BI TO APPLIED AI
FROM BI TO APPLIED AILior Sidi
 
Forget about AI and do Mathematical Modelling instead!
Forget about AI and do Mathematical Modelling instead!Forget about AI and do Mathematical Modelling instead!
Forget about AI and do Mathematical Modelling instead!Florian Wilhelm
 

Similar to Debugging AI (20)

Debugging AI
Debugging AIDebugging AI
Debugging AI
 
The State of Artificial Intelligence in 2018: A Good Old Fashioned Report
The State of Artificial Intelligence in 2018: A Good Old Fashioned ReportThe State of Artificial Intelligence in 2018: A Good Old Fashioned Report
The State of Artificial Intelligence in 2018: A Good Old Fashioned Report
 
State Of AI 2018
State Of AI 2018State Of AI 2018
State Of AI 2018
 
Big Data & Machine Learning - TDC2013 São Paulo - 12/0713
Big Data & Machine Learning - TDC2013 São Paulo - 12/0713Big Data & Machine Learning - TDC2013 São Paulo - 12/0713
Big Data & Machine Learning - TDC2013 São Paulo - 12/0713
 
Big Data & Machine Learning - TDC2013 Sao Paulo
Big Data & Machine Learning - TDC2013 Sao PauloBig Data & Machine Learning - TDC2013 Sao Paulo
Big Data & Machine Learning - TDC2013 Sao Paulo
 
Human-Centered AI: Scalable, Interactive Tools for Interpretation and Attribu...
Human-Centered AI: Scalable, Interactive Tools for Interpretation and Attribu...Human-Centered AI: Scalable, Interactive Tools for Interpretation and Attribu...
Human-Centered AI: Scalable, Interactive Tools for Interpretation and Attribu...
 
UNCOVERING FAKE NEWS BY MEANS OF SOCIAL NETWORK ANALYSIS
UNCOVERING FAKE NEWS BY MEANS OF SOCIAL NETWORK ANALYSISUNCOVERING FAKE NEWS BY MEANS OF SOCIAL NETWORK ANALYSIS
UNCOVERING FAKE NEWS BY MEANS OF SOCIAL NETWORK ANALYSIS
 
UNCOVERING FAKE NEWS BY MEANS OF SOCIAL NETWORK ANALYSIS
UNCOVERING FAKE NEWS BY MEANS OF SOCIAL NETWORK ANALYSISUNCOVERING FAKE NEWS BY MEANS OF SOCIAL NETWORK ANALYSIS
UNCOVERING FAKE NEWS BY MEANS OF SOCIAL NETWORK ANALYSIS
 
Managing machine learning
Managing machine learningManaging machine learning
Managing machine learning
 
AI Orange Belt - Session 2
AI Orange Belt - Session 2AI Orange Belt - Session 2
AI Orange Belt - Session 2
 
20181212 ibm aot
20181212 ibm aot20181212 ibm aot
20181212 ibm aot
 
Explainable AI in Industry (FAT* 2020 Tutorial)
Explainable AI in Industry (FAT* 2020 Tutorial)Explainable AI in Industry (FAT* 2020 Tutorial)
Explainable AI in Industry (FAT* 2020 Tutorial)
 
Applied AI Workshop - Presentation - Connect Day GDL
Applied AI Workshop - Presentation - Connect Day GDLApplied AI Workshop - Presentation - Connect Day GDL
Applied AI Workshop - Presentation - Connect Day GDL
 
Machine learning for sensor Data Analytics
Machine learning for sensor Data AnalyticsMachine learning for sensor Data Analytics
Machine learning for sensor Data Analytics
 
Model governance in the age of data science & AI
Model governance in the age of data science & AIModel governance in the age of data science & AI
Model governance in the age of data science & AI
 
Quant university MRM and machine learning
Quant university MRM and machine learningQuant university MRM and machine learning
Quant university MRM and machine learning
 
Human in the loop: Bayesian Rules Enabling Explainable AI
Human in the loop: Bayesian Rules Enabling Explainable AIHuman in the loop: Bayesian Rules Enabling Explainable AI
Human in the loop: Bayesian Rules Enabling Explainable AI
 
Deep Learning - Hype, Reality and Applications in Manufacturing
Deep Learning - Hype, Reality and Applications in ManufacturingDeep Learning - Hype, Reality and Applications in Manufacturing
Deep Learning - Hype, Reality and Applications in Manufacturing
 
FROM BI TO APPLIED AI
FROM BI TO APPLIED AIFROM BI TO APPLIED AI
FROM BI TO APPLIED AI
 
Forget about AI and do Mathematical Modelling instead!
Forget about AI and do Mathematical Modelling instead!Forget about AI and do Mathematical Modelling instead!
Forget about AI and do Mathematical Modelling instead!
 

Recently uploaded

Re-membering the Bard: Revisiting The Compleat Wrks of Wllm Shkspr (Abridged)...
Re-membering the Bard: Revisiting The Compleat Wrks of Wllm Shkspr (Abridged)...Re-membering the Bard: Revisiting The Compleat Wrks of Wllm Shkspr (Abridged)...
Re-membering the Bard: Revisiting The Compleat Wrks of Wllm Shkspr (Abridged)...Hasting Chen
 
Thirunelveli call girls Tamil escorts 7877702510
Thirunelveli call girls Tamil escorts 7877702510Thirunelveli call girls Tamil escorts 7877702510
Thirunelveli call girls Tamil escorts 7877702510Vipesco
 
No Advance 8868886958 Chandigarh Call Girls , Indian Call Girls For Full Nigh...
No Advance 8868886958 Chandigarh Call Girls , Indian Call Girls For Full Nigh...No Advance 8868886958 Chandigarh Call Girls , Indian Call Girls For Full Nigh...
No Advance 8868886958 Chandigarh Call Girls , Indian Call Girls For Full Nigh...Sheetaleventcompany
 
Andrés Ramírez Gossler, Facundo Schinnea - eCommerce Day Chile 2024
Andrés Ramírez Gossler, Facundo Schinnea - eCommerce Day Chile 2024Andrés Ramírez Gossler, Facundo Schinnea - eCommerce Day Chile 2024
Andrés Ramírez Gossler, Facundo Schinnea - eCommerce Day Chile 2024eCommerce Institute
 
Open Source Strategy in Logistics 2015_Henrik Hankedvz-d-nl-log-conference.pdf
Open Source Strategy in Logistics 2015_Henrik Hankedvz-d-nl-log-conference.pdfOpen Source Strategy in Logistics 2015_Henrik Hankedvz-d-nl-log-conference.pdf
Open Source Strategy in Logistics 2015_Henrik Hankedvz-d-nl-log-conference.pdfhenrik385807
 
SaaStr Workshop Wednesday w: Jason Lemkin, SaaStr
SaaStr Workshop Wednesday w: Jason Lemkin, SaaStrSaaStr Workshop Wednesday w: Jason Lemkin, SaaStr
SaaStr Workshop Wednesday w: Jason Lemkin, SaaStrsaastr
 
Mohammad_Alnahdi_Oral_Presentation_Assignment.pptx
Mohammad_Alnahdi_Oral_Presentation_Assignment.pptxMohammad_Alnahdi_Oral_Presentation_Assignment.pptx
Mohammad_Alnahdi_Oral_Presentation_Assignment.pptxmohammadalnahdi22
 
BDSM⚡Call Girls in Sector 93 Noida Escorts >༒8448380779 Escort Service
BDSM⚡Call Girls in Sector 93 Noida Escorts >༒8448380779 Escort ServiceBDSM⚡Call Girls in Sector 93 Noida Escorts >༒8448380779 Escort Service
BDSM⚡Call Girls in Sector 93 Noida Escorts >༒8448380779 Escort ServiceDelhi Call girls
 
Governance and Nation-Building in Nigeria: Some Reflections on Options for Po...
Governance and Nation-Building in Nigeria: Some Reflections on Options for Po...Governance and Nation-Building in Nigeria: Some Reflections on Options for Po...
Governance and Nation-Building in Nigeria: Some Reflections on Options for Po...Kayode Fayemi
 
Exploring protein-protein interactions by Weak Affinity Chromatography (WAC) ...
Exploring protein-protein interactions by Weak Affinity Chromatography (WAC) ...Exploring protein-protein interactions by Weak Affinity Chromatography (WAC) ...
Exploring protein-protein interactions by Weak Affinity Chromatography (WAC) ...Salam Al-Karadaghi
 
Microsoft Copilot AI for Everyone - created by AI
Microsoft Copilot AI for Everyone - created by AIMicrosoft Copilot AI for Everyone - created by AI
Microsoft Copilot AI for Everyone - created by AITatiana Gurgel
 
CTAC 2024 Valencia - Sven Zoelle - Most Crucial Invest to Digitalisation_slid...
CTAC 2024 Valencia - Sven Zoelle - Most Crucial Invest to Digitalisation_slid...CTAC 2024 Valencia - Sven Zoelle - Most Crucial Invest to Digitalisation_slid...
CTAC 2024 Valencia - Sven Zoelle - Most Crucial Invest to Digitalisation_slid...henrik385807
 
Night 7k Call Girls Noida Sector 128 Call Me: 8448380779
Night 7k Call Girls Noida Sector 128 Call Me: 8448380779Night 7k Call Girls Noida Sector 128 Call Me: 8448380779
Night 7k Call Girls Noida Sector 128 Call Me: 8448380779Delhi Call girls
 
WhatsApp 📞 9892124323 ✅Call Girls In Juhu ( Mumbai )
WhatsApp 📞 9892124323 ✅Call Girls In Juhu ( Mumbai )WhatsApp 📞 9892124323 ✅Call Girls In Juhu ( Mumbai )
WhatsApp 📞 9892124323 ✅Call Girls In Juhu ( Mumbai )Pooja Nehwal
 
Call Girl Number in Khar Mumbai📲 9892124323 💞 Full Night Enjoy
Call Girl Number in Khar Mumbai📲 9892124323 💞 Full Night EnjoyCall Girl Number in Khar Mumbai📲 9892124323 💞 Full Night Enjoy
Call Girl Number in Khar Mumbai📲 9892124323 💞 Full Night EnjoyPooja Nehwal
 
ANCHORING SCRIPT FOR A CULTURAL EVENT.docx
ANCHORING SCRIPT FOR A CULTURAL EVENT.docxANCHORING SCRIPT FOR A CULTURAL EVENT.docx
ANCHORING SCRIPT FOR A CULTURAL EVENT.docxNikitaBankoti2
 
Mathematics of Finance Presentation.pptx
Mathematics of Finance Presentation.pptxMathematics of Finance Presentation.pptx
Mathematics of Finance Presentation.pptxMoumonDas2
 
Presentation on Engagement in Book Clubs
Presentation on Engagement in Book ClubsPresentation on Engagement in Book Clubs
Presentation on Engagement in Book Clubssamaasim06
 
Call Girls in Sarojini Nagar Market Delhi 💯 Call Us 🔝8264348440🔝
Call Girls in Sarojini Nagar Market Delhi 💯 Call Us 🔝8264348440🔝Call Girls in Sarojini Nagar Market Delhi 💯 Call Us 🔝8264348440🔝
Call Girls in Sarojini Nagar Market Delhi 💯 Call Us 🔝8264348440🔝soniya singh
 
Russian Call Girls in Kolkata Vaishnavi 🤌 8250192130 🚀 Vip Call Girls Kolkata
Russian Call Girls in Kolkata Vaishnavi 🤌  8250192130 🚀 Vip Call Girls KolkataRussian Call Girls in Kolkata Vaishnavi 🤌  8250192130 🚀 Vip Call Girls Kolkata
Russian Call Girls in Kolkata Vaishnavi 🤌 8250192130 🚀 Vip Call Girls Kolkataanamikaraghav4
 

Recently uploaded (20)

Re-membering the Bard: Revisiting The Compleat Wrks of Wllm Shkspr (Abridged)...
Re-membering the Bard: Revisiting The Compleat Wrks of Wllm Shkspr (Abridged)...Re-membering the Bard: Revisiting The Compleat Wrks of Wllm Shkspr (Abridged)...
Re-membering the Bard: Revisiting The Compleat Wrks of Wllm Shkspr (Abridged)...
 
Thirunelveli call girls Tamil escorts 7877702510
Thirunelveli call girls Tamil escorts 7877702510Thirunelveli call girls Tamil escorts 7877702510
Thirunelveli call girls Tamil escorts 7877702510
 
No Advance 8868886958 Chandigarh Call Girls , Indian Call Girls For Full Nigh...
No Advance 8868886958 Chandigarh Call Girls , Indian Call Girls For Full Nigh...No Advance 8868886958 Chandigarh Call Girls , Indian Call Girls For Full Nigh...
No Advance 8868886958 Chandigarh Call Girls , Indian Call Girls For Full Nigh...
 
Andrés Ramírez Gossler, Facundo Schinnea - eCommerce Day Chile 2024
Andrés Ramírez Gossler, Facundo Schinnea - eCommerce Day Chile 2024Andrés Ramírez Gossler, Facundo Schinnea - eCommerce Day Chile 2024
Andrés Ramírez Gossler, Facundo Schinnea - eCommerce Day Chile 2024
 
Open Source Strategy in Logistics 2015_Henrik Hankedvz-d-nl-log-conference.pdf
Open Source Strategy in Logistics 2015_Henrik Hankedvz-d-nl-log-conference.pdfOpen Source Strategy in Logistics 2015_Henrik Hankedvz-d-nl-log-conference.pdf
Open Source Strategy in Logistics 2015_Henrik Hankedvz-d-nl-log-conference.pdf
 
SaaStr Workshop Wednesday w: Jason Lemkin, SaaStr
SaaStr Workshop Wednesday w: Jason Lemkin, SaaStrSaaStr Workshop Wednesday w: Jason Lemkin, SaaStr
SaaStr Workshop Wednesday w: Jason Lemkin, SaaStr
 
Mohammad_Alnahdi_Oral_Presentation_Assignment.pptx
Mohammad_Alnahdi_Oral_Presentation_Assignment.pptxMohammad_Alnahdi_Oral_Presentation_Assignment.pptx
Mohammad_Alnahdi_Oral_Presentation_Assignment.pptx
 
BDSM⚡Call Girls in Sector 93 Noida Escorts >༒8448380779 Escort Service
BDSM⚡Call Girls in Sector 93 Noida Escorts >༒8448380779 Escort ServiceBDSM⚡Call Girls in Sector 93 Noida Escorts >༒8448380779 Escort Service
BDSM⚡Call Girls in Sector 93 Noida Escorts >༒8448380779 Escort Service
 
Governance and Nation-Building in Nigeria: Some Reflections on Options for Po...
Governance and Nation-Building in Nigeria: Some Reflections on Options for Po...Governance and Nation-Building in Nigeria: Some Reflections on Options for Po...
Governance and Nation-Building in Nigeria: Some Reflections on Options for Po...
 
Exploring protein-protein interactions by Weak Affinity Chromatography (WAC) ...
Exploring protein-protein interactions by Weak Affinity Chromatography (WAC) ...Exploring protein-protein interactions by Weak Affinity Chromatography (WAC) ...
Exploring protein-protein interactions by Weak Affinity Chromatography (WAC) ...
 
Microsoft Copilot AI for Everyone - created by AI
Microsoft Copilot AI for Everyone - created by AIMicrosoft Copilot AI for Everyone - created by AI
Microsoft Copilot AI for Everyone - created by AI
 
CTAC 2024 Valencia - Sven Zoelle - Most Crucial Invest to Digitalisation_slid...
CTAC 2024 Valencia - Sven Zoelle - Most Crucial Invest to Digitalisation_slid...CTAC 2024 Valencia - Sven Zoelle - Most Crucial Invest to Digitalisation_slid...
CTAC 2024 Valencia - Sven Zoelle - Most Crucial Invest to Digitalisation_slid...
 
Night 7k Call Girls Noida Sector 128 Call Me: 8448380779
Night 7k Call Girls Noida Sector 128 Call Me: 8448380779Night 7k Call Girls Noida Sector 128 Call Me: 8448380779
Night 7k Call Girls Noida Sector 128 Call Me: 8448380779
 
WhatsApp 📞 9892124323 ✅Call Girls In Juhu ( Mumbai )
WhatsApp 📞 9892124323 ✅Call Girls In Juhu ( Mumbai )WhatsApp 📞 9892124323 ✅Call Girls In Juhu ( Mumbai )
WhatsApp 📞 9892124323 ✅Call Girls In Juhu ( Mumbai )
 
Call Girl Number in Khar Mumbai📲 9892124323 💞 Full Night Enjoy
Call Girl Number in Khar Mumbai📲 9892124323 💞 Full Night EnjoyCall Girl Number in Khar Mumbai📲 9892124323 💞 Full Night Enjoy
Call Girl Number in Khar Mumbai📲 9892124323 💞 Full Night Enjoy
 
ANCHORING SCRIPT FOR A CULTURAL EVENT.docx
ANCHORING SCRIPT FOR A CULTURAL EVENT.docxANCHORING SCRIPT FOR A CULTURAL EVENT.docx
ANCHORING SCRIPT FOR A CULTURAL EVENT.docx
 
Mathematics of Finance Presentation.pptx
Mathematics of Finance Presentation.pptxMathematics of Finance Presentation.pptx
Mathematics of Finance Presentation.pptx
 
Presentation on Engagement in Book Clubs
Presentation on Engagement in Book ClubsPresentation on Engagement in Book Clubs
Presentation on Engagement in Book Clubs
 
Call Girls in Sarojini Nagar Market Delhi 💯 Call Us 🔝8264348440🔝
Call Girls in Sarojini Nagar Market Delhi 💯 Call Us 🔝8264348440🔝Call Girls in Sarojini Nagar Market Delhi 💯 Call Us 🔝8264348440🔝
Call Girls in Sarojini Nagar Market Delhi 💯 Call Us 🔝8264348440🔝
 
Russian Call Girls in Kolkata Vaishnavi 🤌 8250192130 🚀 Vip Call Girls Kolkata
Russian Call Girls in Kolkata Vaishnavi 🤌  8250192130 🚀 Vip Call Girls KolkataRussian Call Girls in Kolkata Vaishnavi 🤌  8250192130 🚀 Vip Call Girls Kolkata
Russian Call Girls in Kolkata Vaishnavi 🤌 8250192130 🚀 Vip Call Girls Kolkata
 

Debugging AI

  • 1.
  • 3. 3Photo by Gratisography from Pexels „Debugging AI“ is deliberately ambiguous.
  • 4. What drives the current AI hype? 4
  • 6. 6 More data Photo by Negative Space from Pexels
  • 7. 7 New business opportunities
 (plus attention) By Dllu - Own work, CC BY-SA 4.0, https://commons.wikimedia.org/w/index.php?curid=63450446
  • 8. 8 and language processing. 28 Chapter 2 – Why is AI important? AI is important because, for the first time, traditionally human capabilities can be undertaken in software inexpensively and at scale. AI can be applied to every sector to enable new possibilities and efficiencies. 40 Chapter 3 – Why has AI come of age? Specialised hardware, availability of training data, new algorithms and increased investment, among other factors, have enabled an inflection point in AI capability. After seven false dawns since the 1950s, AI technology has come of age. constraints of human experience 82 Chapter 6 – The war While demand for AI professiona winners and losers are emerging Part 3: The AI Disrup 96 Chapter 7 – Europe’s The landscape for entrepreneurs AI startups are maturing, bringing industries, and navigating new op While the UK is the powerhouse o France may extend their influence https://www.mmcventures.com/wp-content/uploads/2019/02/The-State-of-AI-2019-Divergence.pdf Speed of development
  • 9. https://www.mmcventures.com/wp-content/uploads/2019/02/The-State-of-AI-2019-Divergence.pdf 9 years (Fig. 23), to an estimated $15bn in 2018 (CB Insights, MMC Ventures). Today’s leading technology companies – including Apple, Amazon, Facebook, Google, IBM, Microsoft and Salesforce – are also spending heavily on research and personnel to develop and deploy AI. Internal corporate investment on AI, among just the top 35 high tech and advanced manufacturing companies investing in AI, may be 2.0x to 4.5x greater than the capital invested by venture capital firms, private equity firms and other sources of external funding combined (McKinsey), further catalysing progress. have increased fifteen-fold in five years, to an estimated $15bn in 2018. (CB Insights, MMC Ventures) Source: CB Insights, MMC Ventures Fig 23. Venture capital investment in AI has increased 15-fold in five years 0 200 400 600 800 1000 1200 0 2 4 6 8 10 12 14 16 20132012 2014 2015 2016 2017 2018E Fig. X: Venture capital investment in AI has increased 15-fold in five years AIdeals Disclosed Funding (right axis) Number (left axis) AIdealinvestment($billion) High valuation
  • 10. 10 Fear Photo by samer daboul from Pexels
  • 11. It’s important to understand… 11 that these factors impact your AI project.
  • 12. What is AI? Strong AI vs. Weak AI AI is used as a generic term for a set of tools to cope with a certain set of problems. Machine Learning is a subset of this AI-toolset: „Programming by example“. Other subsets are knowledge representation, planning, reasoning. AI uses probabilistic logic instead of boolean logic. 12
  • 13. 13 „The brown quick fox jumps over the lazy dog“ https://www.mcohen.io/2017/machine-learning-explained-in-three-easy-steps/ brown quick
  • 14. Properties of AI problems Hard to code „by hand“ • Requires non-formalized knowledge (experiential knowledge) • Or even not yet existing knowledge Afflicted with uncertainty (or missing information) Changes rapidly (making it unreasonable to adopt software manually) 14
  • 15. Playing Chess, playing Go 15 f(s) f(s) f(s) f(s) f(s) f(s) f(s) f(s) 1 0 0 1
  • 16. Another typical problem: Computer vision 16
  • 17. This is typically solved using „Deep neural networks“ Neural networks are simple (sort of): 17 ⨁ Input 1 Input 2 Input 3 Output Weight 1 Weight 2 Weight 3
  • 18. Deep Networks combine these simple elements. 18 https://courses.cs.ut.ee/MTAT.03.291/2015_spring/uploads/Main/Artificial%20neural%20network%20for%20image%20classification.pdf complex patterns in the data as it is propagated throught the network. For example, if the network is given a task to recognize a face, the first hidden layer might act as a line detector, the second hidden takes these lines as input and puts them together to form a nose, the third hidden layer takes the nose and matches it with an eye and so on, until finally the whole face is constructed. This hierachy enables the network to eventually recognize very complex objects.3 As stated before, the network is able to accurately approximate an arbitrary function by altering its weights in a systematic way. Initially, the weights are given random values and the network must be trained in order to find the weigth parameters that produce the desired effect. In order to achieve Figure 2. An example of neural network layers as feature detectors.4 bles the network to eventually recognize very complex objects.
  • 19. 19Photo by Gratisography from Pexels digging deeper into „debugging“
  • 20. Recap: Properties of AI problems Hard to code „by hand“ • Requires non-formalized knowledge (experiential knowledge) • Or even not yet existing knowledge Afflicted with uncertainty (or missing information) Changes rapidly (making it unreasonable to adopt software manually) 20
  • 21. 21 These properties make verification hard „by design“ Recap: Properties of AI problems Hard to code „by hand“ • Requires non-formalized knowledge (experiential knowledge) • Or even not yet existing knowledge
  • 22. 22Photo by Pixabay from Pexels Problems with unknown truth For example: Medical classification problem What is the correct diagnosis? What is the correct therapy? You won’t know (maybe until your patient either recovers or dies?) Same is true for customer support systems. Is you customer satisfied? You’ll probably only know by loosing him/her as a customer.
  • 23. 23https://towardsdatascience.com/gender-bias-word-embeddings-76d9806a0e17, Photo by rawpixel.com from Pexels ML replicates bias in the data Example: Conceptual similarities from word embeddings With words typically collocated, you can ask you model for conceptual similarities: king - man + woman ⇾ queen Depending on your input corpus, your model will give you doctor - man + woman ⇾ nurse Do you really run a sexist, racist chatbot on your website?
  • 24. AI systems break fundamental patterns we developed as an industry • No (or very little) isolation. You need to verify and retrain the whole system. • Higher dimension of failure space • Time intense training cycles (instead immediate feedback cycles) • Non-stationary nature of ML systems 24http://ai.stanford.edu/~zayd/why-is-machine-learning-hard.html
  • 26. We need … 26 … new quality management approaches
  • 28. Model requirements Check whether the product / project requirements can be met with the model choices. Example:
 Classify documents Distinguish Cards (ID, Member Cards), structured documents or free text? All of them? Or even mixed? Are you also trying to extract information from these documents? 28
  • 29. Quality management on input data Machine learning is programming by example. Make sure your examples are legit. 29
  • 30. Use BI, data visualization, statistics, Excel Implement quality management for your input data. Visualize, use statistical metrics, use data science on the input data, e.g. using Tableau, Looker, Excel, whatever suits you) E.g., to identify bias in the input data, check whether data is balanced, etc. Use version control on input data (and processing steps) (see dvc.org, that's not perfect, but what is?). Make sure data is versioned alongside code and model. If you create the data from another technical system, manage quality of the generating code. Try to document / check for documentation: Which distribution, what features. 30
  • 31. With supervised learning, you need supervisors. Establish panel of judges both on input data labelling and on outcomes. Due to the non-stationary nature this is is an ongoing task, not a closed project. Crowdsource your data labelling or let clickworkers do their job. QM labelling, again using data visualization, fraud detection (outlier detection, etc.) 31
  • 32. Make sure you know the data Use generated test data with known patterns, because otherwise you won’t know if you miss whole categories. 32
  • 33. Cope with probabilistic logic The world is not black and white, it's full of color. Make sure you see it! 33
  • 34. Work with test sets To test for non-binary outcomes (i.e., results with confidence level), you need to handle test set as opposed to sets of single test outcomes: For example test for outcome confidence distribution. For non-stationary systems: establish test monitoring to accept regression. „Accept new model if result is at least 95% of last model.“ Do not only test outcomes, but implement inspection tools. E.g., in RoboCup Simulation map agent movement paths. 34
  • 35. Work with specific tools These depend on your model requirements 35
  • 36. These techniques are not generally applicable Alter real data to fabricate test data with known properties. Modify input variables and observe model behaviour. Usually small changes should have small effects. Sometimes, you even know the direction, amount or characteristic of effects by applying common sense. Measure at the right granularity (so the system won't utilize loopholes) 36
  • 37. Add a safety net AI is a tool like any other. It is created by humans, and we all make mistakes. 37
  • 38. Add a safety net Implement multi-layer security fallbacks for subsets of your problem (also to be used while testing), like „emergency break systems“. Test these. Use for testing: If you need your security fallback too often, your model may be bad. Implement rules (i.e., code) for parts of your problem. Just in research: Add explainability, local properties by black-box tests on the model to verify the „anchor rules" (https:// homes.cs.washington.edu/~marcotcr/aaai18.pdf) 38
  • 39. © data42 GmbH Speed / cost benefits of Machine Learning often only apply if you accept a non-zero failure rate Think twice before implementing high stakes AI applications (E.g., do NOT use AI to screen applications in HR) 39 26.02.18
  • 40. © data42 GmbH Derive new knowledge from AI … to solve new problems. 40 26.02.18
  • 41. Thank you - and be curious! Dr. Christian Betz, data42