Each week we will share the latest artificial intelligence news and insights, curated for professionals, board members, and entrepreneurs, as we learn to navigate together the opportunities and threats arising from the evolution of AI and technology.
This week in AI
In AI news this week, we review the controversy from a Google associate asserting that chatbot technology is now sentient, federated learning insights and the latest funding news.
Did Google make a sentient AI?
Blake Lemoine, a Google employee, published some of the conversations he had with a chatbot using Google’s Language Model for Dialogue Application (also known as LaMDA) in a Washington Post article last week. He was placed on leave by his employer after Google claimed he violated the confidentiality sections of his employment contract. Lemoine claimed that the conversations were with what he termed a “person”, going as far as to recommend that LaMDA get its own attorney.
Even though LaMDA is extremely convincing, the hurdle for what is termed Generalised AI is still some way off. For example, LaMDA’s ethical challenges were cited back in a January 2022 paper from a group of Google employees and associates, where they wrote, “We study the benefits of model scaling with LaMDA on our three key metrics: quality, safety, and groundedness. We observe that … model scaling alone improves quality, but its improvements on safety and groundedness are far behind human performance”. The inability of the model to manage “safety” means LaMDA’s automated responses are not consistent with what the team termed “a set of human values, such as preventing harmful suggestions and unfair bias”
Bias and safety are critical considerations as AI models scale faster, fooling even sophisticated professionals, with practical and legal safeguards a priority to ensure the responsible use of this awesome technology.
Federate Learning Insights
Federated Learning is an important feature of AI Forum’s upcoming report, The State of AI in Healthcare. AI Forum Advisory Board member, Dr Matthew Crowson, recently contributed to a well received paper on this topic entitled, A systematic review of federated learning applications for biomedical data. From his recent summary of the paper, “Traditionally, if Hospital ‘A’ wanted to engage in collaborative research with Hospital ‘B,’ the institutions would navigate a lengthy administrative process involving shared data use agreements, research contracts, and other logistical hurdles. Successfully navigating these steps can take months and years in some cases.
This is where federated learning comes in. Federated learning is a relatively new collaborative approach that enables institutions to develop machine learning models together while keeping their data within their respective firewalls. Instead of sharing the data itself, the participating institutions train a local version of a model and share the model weights. A central version of the model is updated using these contributed weights. Voila, a collaboratively trained model!”
AI Forum will be publishing more information on the topic of privacy and data sovereignty for Learning Data during the Summer, along with in-depth workshops at the annual conference in the Cayman Islands.
AI Funding News
In funding news, MedTech Innovator published the list of 50 companies selected to participate in the organization’s flagship four-month Accelerator program. The 2022 Accelerator program had a five percent acceptance rate with the top fifty selected from a pool of more than one thousand applications – originating from forty nine countries and forty three U.S. States. An audience vote will determine the winner of the three hundred and fifty thousand US dollar prize at the grand finale on October 24-26 2022, held in Boston, Massachusetts.
In a strong week for Healthcare AI, deals last week included:
- Insilico Medicine, a Hong Kong based clinical-stage artificial intelligence drug discovery company, raising sixty million US dollars in Series D finance
- CHARM Therapeutics, a UK based three dimensional deep learning research company founded raised fifty million US dollars in Series A finance and
- Peptone, a UK based startup using atomic imaging techniques and AI to find new medicines, raised forty million US dollars in Series A funding.
In the finance field, Delphia, based in Toronto, Canada, raised sixty million US dollars in Series A. From their website, “Delphia doesn’t use AI to predict a stock’s price – that’s a fool’s errand. Instead, we use it to predict fundamentals – things like sales or profit – not just today, but quarters ahead. Armed with these predictions, our algorithm goes hunting for where the market is primed for surprise.”
Thanks for watching
That’s all for this week. We hope it was insightful. Please let us know what topics you want us to cover by leaving a comment on YouTube.
Don’t forget to subscribe to our YouTube channel
