Conference Learning Offload

Last week I went to 3, real life and in person conferences. I listened, I spoke and I chaired. It was a feast of learning and here I am taking some time to reflect so that I don’t waste and forget all the goodness that I heard.

The truly eye-popping

It was a rare privilege to hear Alex Bornyakov, the Deputy Prime Minister of Digital Transformation from Ukraine speak. The room was so hushed that we really could have heard a pin drop. He mused on the resilience of an empowered network versus a Tsar based hierarchy and later Mat Ocko talked about the AI being deployed to counter AI in that war zone. It is breathtaking what is technically possible when the “normal” rules of personal privacy, security and commercial liability are suspended. We entered a world of fully autonomous counter drones, continuous earth observation (possibly surveillance) and population wide, real time message translation and correlation. (For less capital investment than a single F35).

Is social media a machine making us easier for algorithms to serve?

As one would expect from someone who delivered the Reith lectures, Stuart Russell made me think about the outcomes of AI in new ways.

• Alan Turing pointed out that we should expect to cede power to the machines, so we should seek to avoid a chess match with them, because we won’t win.
• We cannot expect to be able to specify outcomes for AI in the same way that we cannot foresee all the implications of our actions, think of King Midas. The third wish from the genie in the magic lamp tends to be a wish to undo the previous 2 wishes.
• Algorithms do not know that we exist, they do not understand our plastic future preferences. It is easier for the algorithms to modify our preferences to make them easier to satisfy. (See social media.)

There is hope, but we need fundamental research and the thoughtful policy like that emerging from China, Kai-Fu Lee told us.

What is AI anyway?

Jacomo Corbo asked insightful questions about the challenges of regulating or even understanding AI.

• What do we do if the most useful representation of data is not intelligible?
• Is it enough for AI to learn or must it keep learning?

Collaboration, or not

I had a tussle with Filipe Gracio, does collaboration follow shared data or does data sharing happen as a result of collaboration? I suspect the answer depends on whether you are an engineer or not.

Data discussions usually boil down to trust, but it is incumbent on the service provide to be trustworthy rather than the user to trust them. An important linguistic trick from Aidan Peppin. With that, we need to avoid digital resignation by elevating the power of consent in our communities and ensuring an equitable share of the benefits. They are challenges Kevin Macnish and Milly Zimeta are wrestling with and Sarah Boyd captured with my favourite quote, “change moves at the speed of trust”.

Helen Markides and Kevin Yue gave cracking presentations on data sharing and the challenges of valuation, commercial and governance models, they will be interesting proving grounds.

The power of agency

I learned a great deal about health data, the power of agency and how the answer to an enormous number of problems seems to be patient controlled health records.

• Wouldn’t the world be a better place if we all had training on how to avoid creating records that didn’t compromise privacy?
• As Tim Ferris said, the 4 principal use cases for health data all involve the same information configured in different ways. (Direct patient care, research, planning and care co-ordination.) He was really humble about imposing solutions, preferring to make suggestions to ease interoperability.

And Finally

When I spoke there were 4 audience members. It could have been deeply depressing, but we had 100% engagement and a positive outcome. Thank you Gordon Crick and Dan Rossiter.