Opportunities and Challenges Briefing
Artificial Intelligence
and Machine Learning
Andrew Maier, Timothy Jones
TOPIC #1
Identify potential services or automated
processes for Agencies to better detect spear
phishing attempts and avoid ransomware
incidents.
Identify potential services or automated
processes for Agencies to better detect spear
phishing attempts and avoid ransomware
incidents.
Leverage AI for automated Help Desk
applications. Identify tools and methods for
triaging service requests and inquiries.
TOPIC #2
Leverage AI for automated Help Desk
applications. Identify tools and methods for
triaging service requests and inquiries.
Artificial Intelligence is software
that mimics human behavior so
well as to be indistinguishable from
a person.
Machine learning is software that
improves its own task performance
using training data.
Artificial Intelligence (AI)
Machine Learning (ML)
Timeline
Machine learning
2004
?
Artificial intelligence
2012 2017 20XX
Speech ImageText Cognition
JPMorgan Chase introduced a system for
reviewing commercial loan contracts (COIN);
work that used to take loan officers 360,000
hours/yr can now be done in a few seconds.”
Clear potential
“
Other Examples
● Infosys
○ Systems and Network Management
● BP
○ BP Wellfinder
● Ping An Insurance
○ Claim resolution and settlement
But how useful is it for
government?
● Likely not good for defending against spear phishing.
○ Government isn’t as targeted as the private sector.
○ Multiple people recommended YubiKey (hardware).
● Likely very good for defending against ransomware.
○ ML is well suited to protect against static threats
like viruses, ransomware, and spam detection.
● Help-desks present tradeoffs in the short/long term.
○ Good News: Already work in motion:
■ USA.gov’s AI Project
■ GSA’s AI Working Group
■ USCIS Virtual Asst. “Emma”
■ Contact Center of Excellence
○ Bad News:
■ Automating tier-1 support presents a paradox.
● AI Raises numerous infrastructure considerations.
○ Training Data=> Storage costs
○ Processing => GPU premium expenses
○ Network Traffic/Upgrades
○ “Where are the customers’ yachts?”
● AI might create new opportunities
○ Automated “white glove service” DevOps
○ Automated, predictive outbound communications
to reduce inbound calls.
● Workforce disruption -“Human-Machine Relations”
● Data
● Privacy
● Compliance
Governance issues
Workforce Disruption
● What happens to the existing people?
● How to integrate remaining people and machines?
○ “Manufacturing Safety”
■ There’s no shut-off switch.
● What are the ethics of AI?
○ New Google Deepmind Ethics Group
● No Asimov’s 3 Laws of Robotics
Data Issues
● Naked Algorithms vs. Training Data
● Quality
○ Need for “Failure” training data
● Scarcity
● Ownership
● Fragmentation
○ Remember “Islands of Information”?
Privacy
● Regulatory boundaries
○ Financial
○ Medical/Health (HIPAA)
○ Ethnographic
● Opt-In rules
○ “Should I have to agree to being AI’ed, or is the
assumption that I don’t and have to Opt-out”
○ Shanghai CCTV
Compliance
● There’s no auditability
● There’s no reporting
● There’s no transparency
○ “Where’s the console?”
● No AI stack interoperability
○ Beware balkanization of AI Stack
Summary and Recommendations
● Cautious optimism. Test new capabilities in pilots.
● Conduct governance pilots in parallel with AI/ML pilots.
● Leverage existing features (such as Salesforce’s “Einstein”).
● Be a “Fast Follower” to private sector efforts.
● Periodic updates from work on the ground.
Discussion/Q&A

AI and Machine Learning in Government Briefing

  • 1.
    Opportunities and ChallengesBriefing Artificial Intelligence and Machine Learning Andrew Maier, Timothy Jones
  • 2.
    TOPIC #1 Identify potentialservices or automated processes for Agencies to better detect spear phishing attempts and avoid ransomware incidents. Identify potential services or automated processes for Agencies to better detect spear phishing attempts and avoid ransomware incidents.
  • 3.
    Leverage AI forautomated Help Desk applications. Identify tools and methods for triaging service requests and inquiries. TOPIC #2 Leverage AI for automated Help Desk applications. Identify tools and methods for triaging service requests and inquiries.
  • 4.
    Artificial Intelligence issoftware that mimics human behavior so well as to be indistinguishable from a person.
  • 5.
    Machine learning issoftware that improves its own task performance using training data.
  • 6.
  • 7.
  • 8.
    JPMorgan Chase introduceda system for reviewing commercial loan contracts (COIN); work that used to take loan officers 360,000 hours/yr can now be done in a few seconds.” Clear potential “
  • 9.
    Other Examples ● Infosys ○Systems and Network Management ● BP ○ BP Wellfinder ● Ping An Insurance ○ Claim resolution and settlement
  • 10.
    But how usefulis it for government?
  • 11.
    ● Likely notgood for defending against spear phishing. ○ Government isn’t as targeted as the private sector. ○ Multiple people recommended YubiKey (hardware). ● Likely very good for defending against ransomware. ○ ML is well suited to protect against static threats like viruses, ransomware, and spam detection.
  • 12.
    ● Help-desks presenttradeoffs in the short/long term. ○ Good News: Already work in motion: ■ USA.gov’s AI Project ■ GSA’s AI Working Group ■ USCIS Virtual Asst. “Emma” ■ Contact Center of Excellence ○ Bad News: ■ Automating tier-1 support presents a paradox.
  • 13.
    ● AI Raisesnumerous infrastructure considerations. ○ Training Data=> Storage costs ○ Processing => GPU premium expenses ○ Network Traffic/Upgrades ○ “Where are the customers’ yachts?” ● AI might create new opportunities ○ Automated “white glove service” DevOps ○ Automated, predictive outbound communications to reduce inbound calls.
  • 14.
    ● Workforce disruption-“Human-Machine Relations” ● Data ● Privacy ● Compliance Governance issues
  • 15.
    Workforce Disruption ● Whathappens to the existing people? ● How to integrate remaining people and machines? ○ “Manufacturing Safety” ■ There’s no shut-off switch. ● What are the ethics of AI? ○ New Google Deepmind Ethics Group ● No Asimov’s 3 Laws of Robotics
  • 16.
    Data Issues ● NakedAlgorithms vs. Training Data ● Quality ○ Need for “Failure” training data ● Scarcity ● Ownership ● Fragmentation ○ Remember “Islands of Information”?
  • 17.
    Privacy ● Regulatory boundaries ○Financial ○ Medical/Health (HIPAA) ○ Ethnographic ● Opt-In rules ○ “Should I have to agree to being AI’ed, or is the assumption that I don’t and have to Opt-out” ○ Shanghai CCTV
  • 18.
    Compliance ● There’s noauditability ● There’s no reporting ● There’s no transparency ○ “Where’s the console?” ● No AI stack interoperability ○ Beware balkanization of AI Stack
  • 19.
    Summary and Recommendations ●Cautious optimism. Test new capabilities in pilots. ● Conduct governance pilots in parallel with AI/ML pilots. ● Leverage existing features (such as Salesforce’s “Einstein”). ● Be a “Fast Follower” to private sector efforts. ● Periodic updates from work on the ground.
  • 20.