As an attendee at my first Artificial Intelligence (AI) conference, I had few expectations. However, after three days, something didn’t feel right. My limited human brain was puzzled, because all the speakers were human! At a leading-edge AI conference, shouldn’t there be AI speakers and panelists?
That bit aside, there were some cool takeaways:
- There has been $10 billion-$40 billion of recent investment in global AI markets, creating 875 companies based in the U.S., 500 in Canada and 400 each in Israel and China
- There are only 10,000 AI experts worldwide, making for more than $1 million invested per expert. The average AI expert salary at Google DeepMind is $350,000.
- The experts predict that the winners in the space will be explainable (transparent), unsupervised AI solutions. Regulated environments require explanations for behavior. European Union law requires that decisions made by AI be explainable.
- The best AI solutions combine natural intelligence and AI working together–for example, a human driving a car using an AI GPS. Human and AI combined teams are the best chess and Go players in the world, beating both individual AI players and individual humans.
- Through 2027, 17% of all jobs will be replaced by AI solutions. But 10% of new job growth will involve working with AI. This leaves a net 7% job loss, with 10 million jobs disappearing over the next 10 years, more than the number lost in a recession.
- The breakout of effort to build an AI solution is 10% algorithm work, 20% data preparation and 70% AI training.
- AI solutions can pass CAPTCHA tests and score ~500-800 on SAT tests.
- What we call AI experts are really regressive thinking experts, because they must use regressive data to train an AI solution.
Heath Terry from Goldman Sachs noted that with $10 billion -$40 billion being invested into this new market, the winner could be the size of Amazon, Apple, Facebook, Google or Microsoft. Other speakers noted that those mega cap companies were already winning the battle for top AI talent, and would lead this field themselves.
Some overly enthusiastic, Kool Aid-drinking speakers said everyone should be starting an AI project. Although that may not make sense, many interesting AI use cases were covered:
- Replacing seasonal workers at a call center–no need to retrain workers each season
- Give AI an objective, such as sell a surplus of inventory, letting it figure out the details–cross sell, etc.
- Driver assistance–GPS, collision warning, ride sharing
- Medical imaging–for example, spotting cancer through image analysis
- Using voice detection to spot physical disorders and recommend therapy
- Using videos of movement to spot physical disorders and recommend therapy
- Increasing evidence-based clinical interventions–today only 10-20% of these are based on randomized control trials
- Suicide prevention through social media analysis
- Governance of cross border data transfers
- Salesperson replacement by watching top salespeople to see and train AI systems on how they behave
Of note, AI experts working in healthcare diagnostics struggle with using Electronic Health Records (EHR) data, because major EHR providers focus on data capture for billing, not research purposes.
Here are some emerging AI companies to check out further:
- Flamingo AI: Online customer sales and service
- DialogFlow: Conversational interface company acquired by Google which predicts that by 2020, 85% of customer interactions will be through conversational AIs
- H2O.ai: Machine learning platform used by 222 companies in the Fortune 500
- Kogentix: Workbench to simplify big data and AI integration
- deepsense.ai: Machine learning lab
- Veritone: Publicly traded company integrating thousands of AI engines to optimize results
- Day Zero Diagnostics: Identifying bacteria strains that cause hospital infections
- Digital Reasoning: AI for financial services compliance
- Jibo: A social robot as a home companions
- Algomus: AI enabled Virtual Business Analyst
…and some recommended books:
- Life 3.0: Being Human in the Age of Artificial Intelligence by Max Tegmark
- Army of None: The Future of War by Paul Scharre (available April 2018)
My final thought from the conference was “How does one upgrade an AI solution?” The answer partly involves techniques using capsule networks and dynamic variable modeling; however, upgrading AI is still considered to be immature technology with more research needed.
I plan to return to AI World in 2018 if they have non-human AI speakers, and will look forward to hearing from the upgraded versions of those AI speakers in 2019 and beyond.