The Infosys Q1 FY20 Results Press Conference

The AI Bubble

Gerbert & Shira of the Boston Consulting Group, @BCG write in the MIT Sloan Management Review @mitsmr that 'yes, today’s fascination with all things #AI has most of the trappings of a financial #bubble...in most cases there is no clear path for (AI startup) companies to become profitable.' Under the circumstances, small start-ups, having developed their technology, or having run out of venture capital money, will seek to be acquired by larger companies. These acquiring companies, they warn, should be wary. Two things may be worth noting when assessing AI startup companies - first, most algorithms (machine learning especially) used in AI applications are several decades, if not half a century old (a version of backpropagation was, for example, used in the Apollo-11 moon landings). The main new development has been in the development of faster computers on which machine learning algorithms can run, and in the availability of large new data sets on which neural nets/machine learning algorithms can be trained. 

But the first development is ironic in a way - in that, neural nets were intended to be archetypes of parallel computer architectures, so the fact that they are simulated in conventional (though very fast) computers of the conventional type should give one some pause. To be sure, there are some new special-purpose hardware architectures (including new chips) which are designed to optimize the computational power used by neural networks. But this has not yet become altogether commonplace. But even when this comes to become more common, rare will be the startup that owns the data which it will use to train its neural network application. The lack of clear ownership of data means that a startup may not be able to optimize and customize its application to the ultimate user, while the machine learning algorithms themselves are publicly available practically, or actually, for free. Thus, on the critical axis of value creation: the trained network, a crucial aspect - the data used - is not owned by the startup. How then can the typical AI startup aim for profitability? What will be its critical determinant of value and distinguishing characteristic? These questions are precisely the ones that acquiring companies will need to ask themselves and the startups they hope to acquire.

While of course one can see the bubble emerging in AI startups, and AI applications, this bubble may not ultimately prove as harmful as a purely financial bubble (eg the 2008 Global Financial Crisis) did. Some good may come even out of AI investments in companies that may ultimately have to be wound up before turning profitable. Some knowledge generation and diffusion will indeed occur, perhaps some patents will be filed or even approved, and ultimately some value may accrue to the investor, though perhaps not at the scale originally envisaged

The Limits of Ethical AI

.@Joi on 'The Limits of Ethical AI' (which he concedes more properly is 'Limits of Algorithmic Fairness in Actuarial Practice'). On the pervasiveness of AI #Hype: AI has gone 4m being #WhatWeCannotDo (Yet) to being #WhatWeAreMostDefinitelyDoing (Right Now). Martha Minow begins by saying that she cannot think of anyone better than the moderator (Prof Sheila Sen Jassanoff) to bring politics, journalism, philosophy, psychology & empiricism together, adding  ‘that’s probably the most important thing I’m going to say’. But then goes on to make a series of very powerful contributions to the discussion. I don't want to excerpt for fear of quoting w/o context... except an irresistible line 'We are in the year 1900 when it comes to law...' (And @yes_VY, she mentions @JuliaAngwin's work around 1:10:45)

— Satyen Baindur (@Satyen_Baindur) March 8, 2019

https://vimeo.com/311460280 

The AI Hype and VC Funding for Startups

.@ft According to this article from the Financial Times, as many as 40% of 2830 'AI startups' in Europe (@MMC_Ventures Survey) don't use any AI programs at all, in their products (as of the time the survey was taken).  The same survey said that the median size of a funding round for an 'AI startup' in 2018 was 15% larger than a similar funding round was for 'just software' startups. One can see how big an incentive this creates for a 'just software' startup to declare itself an 'AI Startup'. Given this, one is actually a bit surprized that only ~8% of all 2018 startups were AI startups! (though, to be sure, the universe here includes non-software startups too.) But there sure is a lot of #AI #Hype via @IanHathaway https://t.co/2qscQS8PeF

— Satyen Baindur (@Satyen_Baindur) March 5, 2019

What is the Added Business Value of Deploying AI?

An extremely important question in these days of extreme 'AI Hype' is - What is the added business value of deploying AI (this is temporarily setting aside the huge assortment of technical and implementation-related issues, which need their own blog post(s)...). The added business value of AI is more subtle than, for example, the  impact of automation (which can be AI-enabled) that might happen via productivity enhancements, or via increases in revenue, or both in some combination.

AI has an impact on improved decision-making processes, more accurate demand forecasts, or, at a manufacturing level,  the prediction of mechanical failure in machinery - are often indirect, so one needs more subtle and innovative measures of the return on investment (#ROI) in #AI! https://t.co/t0jNT7RIoe

— Satyen Baindur (@Satyen_Baindur) January 30, 2019

AI at Davos 2019

'AI, AI, #AI. It was almost impossible to walk into any room in #Davos2019 and not hear someone talking about artificial intelligence in all its forms...A good general rule right now is that whenever a conversation about AI comes up, a conversation about jobs is sure to follow.' https://t.co/d9B9LUcLZ8

— Satyen Baindur (@Satyen_Baindur) January 30, 2019

Infusing Physics into Machine Learning / AI Algorithms - Animashree Anandkumar