Last year, at CogX 2018, I was blown away by a couple of sessions, most notably one about the New Radical Empiricism, which I wrote about here. I was therefore quite looking forward to CogX 2019. After all, it’s only at an extensive event about AI and Cognitive technologies, that you would see a session titled ‘Understanding Understanding’. And although the weather and some of the organisation left much to be desired, at the end of 3 days and about 40 pages of scribbles, it turns out there plenty to ruminate on this time as well. So here’s a list of distilled thoughts — a few interesting things that caught my imagination and made me think.
The first 5 are broad observations about AI, organisations and challenges for everybody:
- Stuart Russell, Professor of Computer Science at Berkley, made the very interesting point that while the world is going crazy about data, especially to train algorithms. Yet, the whole point of better algorithms is that they will need less data for training. So the data gold-rush may not last forever. You heard it here second.
- Ismail Salim, Canadian serial entrepreneur and author of Exponential Organisations, suggested that we are experiencing the equivalent of 20 simultaneous Gutenberg moments. He made the point that institutions of today are not designed for purpose. He meant it not just for businesses but also for social institutions — including marriage and religion. After all, if people live longer and don’t die, how do you sell religion? (See Jose Cordeiro, below)
- Manuela Veloso runs an AI research team at JP Morgan where she has been running ‘Project Mondrian’. This is a project where an AI engine looks at hundreds/ thousands of charts which are visualisations of prices leading to buy/ sell recommendations based on the underlying mathematics. Project Mondrian explores whether the AI would deliver an accurate recommendation based on just reading the visual charts, with no access to the maths, and has already achieved a 95% accuracy. Of course, AI uses maths, it’s just a different kind of maths to that which currently drives trading decisions.
- Ashok Vaswani is always worth listening to, and he talked about chaining the culture around technology, within the organisation. “Technology is too important to be left to technologists”. He also called out correctly that technology can no longer be just a ‘horizontal’ in the business and it starts to blend other functions such as compliance, reputation and marketing. It’s also key to prompting the question ‘what business are you in?’
- Given the very active discussions around Ethics across the field of AI, it feels like there is also a need for the software (especially AI) industry to create some form of self-regulation. This could be by formal methods — similar to civil engineering, where specialists review and approve plans for bridges or buildings. Or it could be through a code of conduct — similar to a Hippocratic oath that medical professionals sign up to. Given the seriousness of the areas software and AI are being applied to, it should probably be a combination of both, and supported by legislation.
Thanks to my current interest in the ageing society and the problems and opportunities around it, I spent a lot of time in the healthcare stage. Here are 5 interesting takeaways from there:
- Jose Cordeiro, transhumanist and Author of the Spanish book ‘La Muerte De La Muerte’ (or The Death of Death), mounted the most passionate argument for eternal life being within reach of humans. The argument, also supported by some others is that when you look at health at a cellular and chemical level rather than at a disease or organ system level, the game changes. There is a body of people who believe that by 2029 we will have reached a stage where we can add a year of life expectancy each year, and by 2045, we will have defeated death and presumably, ageing. While it’s hard to go wholeheartedly with this premise, there is clearly enough science to suggest that we are increasing lifespans consistently, albeit life expectancy has actually fallen in the UK this year for the first time in many years.
- Dr Jordan Shlain spoke excellently and engagingly about the need for a new “Hippocratic oath 2.0” — this includes building radical transparency, use of AI to drive ethical practice, and restricting data sharing with non-essential providers with patients able to opt out. Shlain also made the valid point that engineers, data scientists, and business leaders in healthcare need to take the Hippocratic oath, not just clinicians.
- Using AI to solve Blood Supply Chain Challenges: The supply of blood for transplants and infusions is more complex than you think. NHS England maintains 1 day of supply on average. Blood has a shelf life of 7 days. But from a single unit of donated blood, there can be up to 128 different ‘products’ created — by type, platelet count etc. Almost 250 hospitals need blood daily, and their demand is highly volatile — anything between 7 and 80 units per day. As you can imagine, this is a highly reactive environment and beset with wastage and ad hoc transport requirements. Kortical has been working with the NHS Blood & Transplant (NHSBT) to bring ad hoc transport and wastage costs down by 50% each. The first round of pilots has brought it down by 15% and a second round will be done including hospital data which will look to integrate hospital data — which will provide better demand predictability.
- NHSx: A new organisation has been set up by the NHS and the Department of Health and Social Care. NHSx will be the face of technology adoption and data management, across areas as diverse as strategy, transparency, agility and standards. All of this with a view to putting Patient experience at the heart of the technology challenge. Experts both within and outside the NHS recognise clearly that the patient experience is broken, data is fragmented, and technology is uneven, and largely not fit for the 21st century. Let's hope NHSx can bring the transformation.
- Ophthalmology is another area which is being changed by AI. Apparently, the highest percentage of outpatients (10%) come in for eye related issues. @Pearsekeane talked through the example of how neural networks were used to classify and diagnose problems. And how 8 skilled ophthalmologists and optometrists performed against the AI system. (A couple of them came close, but none did better). Importantly, the AI did not miss any urgent cases. Note: this has not yet been scaled, it has now gone through POC, and we should watch this space.
There you have it, this is just a starting list, and I could easily make another list of 10. And that’s just a fraction of what was discussed based only on the sessions I attended. What did you see that was noteworthy? And what are your biggest AI challenges?
(I work as the Digital Evangelist for Tata Consultancy Services. All views here are my own.)
