Is AI the answer to all tech problems, or is it part of the problem?

The whole issue of AI working its way into every walk of technology life, frankly worries us at Faultline Online Reporter. According to FinTech researcher CB Insights, some 658 funding deal around AI took place in 2016, putting over $5 billion in VC funds on the line. In total since 2012 some 1,928 AI investments have soaked up over $12.4 billion.

It has crept into our world too, with some providers professing machine learning sits under their recommendation systems, and some video analytics offerings are also said to use machine learning to assess complex video delivery situations. None of them however have a product being used in anger, which you can buy at a list price, today.

For anyone who lived through the dotcom boom and bust, there is that familiar feeling that nothing much has changed, but everyone now uses AI as the buzzword, whether they understand it or not.

It seems that whenever we ask for some concrete examples of a return on investment, we get guided back time and time again to General Electric and its AI driven ability to predict when one of the machines it makes or finances will go wrong. And that’s about it. GE has given several speeches about how much return on investment this has given it, but has not revealed the details.

All of this led us to write a paper on waves of investment funding entitled Transformation Economics, in which we suggested that that $12.4 billion is a fraction of the money that will be spent going forwards, keeping those AI investments alive over the next 5 to 10 years. These moves threaten to pull technology investment out of shape – if you want a successful investment, better suggest that it is driven my machine learning or natural language, otherwise you will not be heard. But if that turns out to be a damp squib, then all those investments will go south.

This week Google has siphoned off its own investment vehicle for AI and pushed 4 of its existing investments into it. The vehicle is called Gradient Ventures, and it will provide capital, resources and education to AI-first startups. Is that as opposed to AI-second start ups?

Google has not said how big the fund is and it is clearly a way of making bets in AI, which we expect will eventually be taken off the balance sheet, and used purely to best use core Google AI.

Two of them Algorithmia and Cape, we already know the extent of Google’s initial investment – Google took part in a $10.5 million round at Algorithmia and spent about $12 million in Cape. If we assume the other two, Cogniac and Aurima, have cost Google a similar amount, then its investment level is a shade over $40 million – nothing for Google to sweat about. But let’s consider the dynamic. If Google begins to invest in companies which can use its Tensorflow technology to best effect, guiding them directly on how best to use it, then its involvement is not purely as an investor – but as a way of promoting its technology. It’s more marketing than investment.

If IBM, Microsoft, Amazon and Facebook and others in the AI arms race do the same, all of them are cutting in on the investment eco-system which has grown naturally around AI, and they too will play favorites.

More companies will seek Google help, as it will look like an investor who cannot fail. Then what happens to those 1,928 other AI investments out there? Are they all destined to become part of one of the main natural language players, and is a sale to Google or another large AI concern the main exit route? If it is there may not be the appetite to make more than 20 to 30 companies in this sector successful – which is bad news for the 1,900 other start-ups. But to stimulate an IPO market in AI, you will need clear winners, with high revenue growth, offering a clear and large ROI for its clients.

If the other 1,900 investments were to fail, there would not be any returns on investment to plow back into next generation tech investments – even a second round of AI investments, so that, at all costs must be avoided. The condition for further investment in a particular company is always that the company has become a little more valuable, so that the next round goes in at a higher valuation. Failure to achieve that will mean the investment is not working on a return on investment basis. Companies like Google can afford to invest on that basis, but pure VC investor funds cannot.

A fall in value at any given AI company on a subsequent investment round is usually the trigger for the VCs to use “price protection” to take over a lot more of the stock and at that point they have absolute control of the company’s destiny. But as the clock ticks on AI investments and little evidence emerges of measurable returns on investment, this is more and more likely to happen. It is probably happening all the time, though not widely advertised.

The defense against a fall in value for the VC fund, is for multiple AI visions to be rolled up into one – 2 or 3 AI startups, which have not proved their ROI, can merge. But the outcome are 2 or 3 visions of how to make money, or to build products, and VCs usually make sure that only one management team survives – the one closest to providing a return.

So given that Google already has Google Ventures and Capital G, which operate as independent funds, and which are already piled into AI, the only reason for getting involved with a new fund that works with Tensorflow is to push that technology. This is marketing then, not investment and we should not confuse the two.

Algorithmia is a community which creates microservices for different AI outputs, such as face detection, analysis of the social buzz and nudity detection for web services; Cogniac provides visual software such as machine vision and defect detection, safety compliance in the workplace, threat analysis, content censorship; while Cape virtualizes drone hardware enabling people to fly drones remotely and Aurima says it is building a multi-sensor deep-learning awareness platform, whatever that is but it looks like IoT.