Explosion AI is a digital studio specializing in AI and natural language processing, emphasizing the importance of high-quality training data for effective machine learning. The document discusses the challenges in data collection, especially in terms of human annotators' experiences and incentives, and proposes solutions like active learning to improve data quality. It concludes that AI systems are not just complex algorithms but rely heavily on well-annotated data and require both read and write access for optimal performance.

1. Teaching AI about
human knowledge
Ines Montani
Explosion AI

2. Explosion AI is a digital studio
specialising in Artificial Intelligence
and Natural Language Processing.
Open-source library for industrial-strength
Natural Language Processing
spaCy’s next-generation Machine Learning library
for deep learning with text
Data Store
coming soon: pre-trained, customisable models
for a variety of languages and domains

3. Machine learning is
programming by
example.

4. Examples are your source
code, training is compilation.
draw examples from the same distribution as
the runtime inputs
goal: system’s prediction given some input
matches label a human would have assigned
examples
training
predictioninput
labels

5. How machines “learn”
def train_tagger(examples):
W = defaultdict(lambda: zeros(n_tags))
for (word, prev, next), human_tag in examples:
scores = W[word] + W[prev] + W[next]
guess = scores.argmax()
if guess != human_tag:
for feat in (word, prev, next):
W[feat][guess] -= 1
W[feat][human_tag] += 1
the weights we'll train
if the guess was wrong, adjust weights
get the best-scoring tag
score each tag given weights & context
examples = words, tags, contexts
decrease score for bad tag in this context
increase score for good tag in this context
Example: training a simple part-of-speech tagger with the perceptron algorithm
(teach the model to recognise verbs, nouns, etc.)

6. The bottleneck in AI
is data, not algorithms.

7. Algorithms are general,
training data is specific.
data quality, data quantity and accuracy
problems are still the biggest problems in AI
(Source: The State of AI survey)
you can extract knowledge from all kinds of
sources, e.g. sentiment from emoji on Reddit 😍
you usually need at least some data specific to
your problem annotated by humans

8. Where human knowledge
in AI really comes from
Images: Amazon Mechanical Turk, depressing.org
Mechanical Turk
human annotators
~$5 per hour
boring tasks
low incentives

9. Don’t expect great data if
you’re boring the shit out of
underpaid people.
Why are we “designing around” this?
“Taking a HIT: Designing around Rejection, Mistrust, Risk, and
Workers’ Experiences in Amazon Mechanical Turk”
(McInnis et al., 2016)
data collection needs the same treatment as all
other human-facing processes
good UX + purpose + incentives = better quality

10. UX-driven data collection
with active learning
assist human with good UX and task structure
the things that are hard for the computer are
usually easy for the human, and vice versa
don’t waste time on what the model already
knows, ask human about what the model is
most interested in
SOLUTION #1

11. model human
tasks
human annotates
all tasks
annotated tasks
are used as training
data for model
Batch learning vs. active learning approach to annotation and training
BATCH
BATCH

12. model human
tasks
human annotates
all tasks
annotated tasks
are used as training
data for model
model chooses
one task
human annotates
chosen task
annotated single task influences model’s
decision on what to ask next
Batch learning vs. active learning approach to annotation and training
BATCH
BATCH
ACTIVE

13. Import knowledge with
pre-trained models
start off with general information about the
language, the world etc.
fine-tune and improve to fit custom needs
big models can work with little training data
backpropagate error signals to correct model
SOLUTION #2

14. user input
word meanings
entity labels
phrase meanings
intent
your examples
“whats the best way to catalinas”
fit meaning
representations
to your data
Backpropagation

15. user input
word meanings
entity labels
phrase meanings
intent
your examples
“whats the best way to catalinas”
fit meaning
representations
to your data
add workaround
for read-only model
Backpropagation

16. If you don’t own the
weights of your model,
you’re missing the
point of deep learning.

17. Deep learning is about
reading and writing.
deep learning = learning internal
representations to help solve your task.
backpropagate errors on the task you care
about to adjust the internal representations
SaaS is great for read-only software – but deep
learning needs read and write access

18. 1. AI systems today are no magically smart
black boxes. They’re made of annotations and
statistics.
2. AI has an HR problem. To improve the
technology, we need to fix the way we collect
data from humans.
3. AI needs write access. This challenges the
current trends towards thin clients and
centralised cloud computing.

19. Thanks!
💥 Explosion AI
explosion.ai
📲 Follow us on Twitter
@explosion_ai
@_inesmontani
@honnibal