AI Talk: Trustworthiness, Watson woes, State of AI

March 5, 2021 / By V. “Juggy” Jagannathan, PhD

This week’s AI Talk…

Trustworthiness of AI in health care

The Consumer Technology Association has just released a new ANSI Standard ANSI/CTA-2090: “The Use of Artificial Intelligence in Health Care: Trustworthiness”. The standard details requirements for health care systems using AI with the goal of engendering trust. The association categorizes three types of trust: human trust, technical trust and regulatory trust.

Human trust revolves around promoting confidence in the use of the systems by its users—physicians, caregivers, health care personnel and of course, end users and patients. So, how is this accomplished? Engineer systems that are easy to use and making them explainable, so users are convinced to perform certain actions when recommended by the system. Just as automobile assistive technologies can go from level 0 (no assistance) to level 5 (completely autonomous), one can envision different ranges of capabilities for AI systems in health care. The bar certainly is higher for systems that focus on autonomous treatment vs. those that are assistive in nature.

Technical trust is focused on ensuring that a system has the right provenance of data, with no introduction of bias, etc.

Regulatory trust ensures that the system follows all appropriate regulations. For instance, the FDA regulates any use of AI for treatment purposes using Software as Medical Device (SaMD) guidelines. The FTC has a set of regulations regarding consumer protection and our Tort laws provide a range of protections.

The requirements spelled out in the new standard have overlapping implications—for example, technical trust and regulatory trust have a direct impact on human trust. It’s a good set of requirements and a great start. However, we still have a long way to go in defining and developing safe, unbiased, ethical and explainable AI.

IBM Watson woes

A decade ago, IBM Watson made a huge splash by winning the Jeopardy game show. It was a triumph of tech using AI and it was truly impressive. After the win, IBM decided to focus its AI might on addressing problems in health care and started a new division: Watson Health. They also developed solutions in oncology and other areas. Unfortunately, the results reported after a trial of its oncology solutions were less than stellar. That’s to be expected—health care is complex and datasets that are needed to train AI systems are diverse and unintegrated. In this case, marketing got ahead of reality and lead to disappointment. Now, IBM is looking at divesting this division. This is a cautionary tale for any company trying to deploy AI solutions, particularly in health care.

State of AI

Want to know what’s happening to the field of AI? Here is a presentation that does justice to that broad theme, offering a comprehensive look at the state of AI. The presentation authors are from the UK investing community and this is the third year in a row they have done such an assessment.

The presentation is packed with useful information across the board. 177 slides with information segmented into the following sections: Introduction, Research, Talent, Industry, Politics and Predictions. Here are a few highlights:

• Transformer models are dominating text understanding applications
• AI techniques are making huge inroads into biology and drug discovery
• Talent is migrating to industry from academia
• AI ethics is driving regulations

The presentation ends with a few predictions for 2021:

• A 10-trillion parameter language model (current record holder is GPT-3 175 billion model)
• A drug-discovery company will be valued at $1 billon+
• Facebook will make major inroads into virtual and augmented reality

We’ll need to tune in next year to see how the authors did with their predictions. This is an excellent compilation – quite technical though and not for the faint of heart!

Acknowledgements

My colleague and friend Brian Ellenberger sent me the links to the Watson story. My half-a-century-old friend and classmate Anandan sent me the link to the state of AI slide deck.

I am always looking for feedback and if you would like me to cover a story, please let me know. “See something, say something!” Leave me a comment below or ask a question on my blogger profile page.

V. “Juggy” Jagannathan, PhD, is Director of Research for 3M M*Modal and is an AI Evangelist with four decades of experience in AI and Computer Science research.

Listen to Juggy Jagannathan discuss AI on the ACDIS podcast.