Silicon Valley loves a new fad. To judge by the spate of fundraising by start-ups in recent weeks, it has found one in an idea that is more than half a century old: artificial intelligence.

“This is the hot place to be at the moment,” says Stephen Purpura, whose own AI company, Context Relevant, has raised more than $44m since it was founded in 2012. By his reckoning, more than 170 start-ups have jumped on the AI bandwagon.

The newcomers to AI believe that the technology has finally caught up with the hopes, bringing a heightened level of intelligence to computers. They promise a new way for humans to interact with machines — and for the machines to encroach on the world of humans in unexpected ways.

“Technologically, it’s a paradigm shift from putting commands into a box to a time when computers watch you and learn,” says Daniel Nadler, another of the AI hopefuls. His company, Kensho, raised $15m recently in pursuit of an ambitious goal: to train computers to replace expensive white-collar workers such as financial analysts.

“We don’t describe what we’re doing as AI — we call it, ‘automating human-intensive knowledge work’,” he says.

The herd mentality of investors partly explains why the AI stampede has become one of the hottest trends in start-up investing since the “big data” slogan launched a thousand entrepreneurial dreams. The investment rounds are still small, reflecting the early stage of most of the start-ups. But the large number of companies raising money and the range of investors point to the extent of the interest.

Besides some of Silicon Valley’s leading venture capital firms — including Khosla Ventures and Greylock Partners — and tech luminaries such as Elon Musk and Peter Thiel, some of the most active backers of AI also include companies which see uses for the technology in their own industries, such as Goldman Sachs.

Every venture capital portfolio now needs its share of investments in the field, says Mr Nadler: the investors who put money into VC funds, known as limited partners, all want to think that they have a stake in the industry’s latest “next big thing”.

The latest AI dawn owes much to new programming techniques for approximating “intelligence” in machines. Foremost among these is machine learning, which involves training machines to identify patterns and make predictions by crunching vast amounts of data. But like other promising new ideas that inspire a rash of start-ups, there is a risk that many companies drawn to the field will struggle to find profitable uses for the technology.

“A lot of these AI platforms are like Swiss army knives,” says Tim Tuttle, chief executive of Expect Labs, which recently raised $13m. “They can do a lot of things, but it’s not clear what the high-value ones will be.”

The result, he says, is a “wild-west mentality” in the industry, as entrepreneurs race to apply AI to every computing problem they can think of.

“I don’t think machine learning, as a standalone technology, is a valuable business,” adds Mr Purpura. “A lot of these things will get acquired.”

The hope that AI will be more than just another passing tech fad is based on its broader potential. Like “big data”, the phrase refers not just to a single technology or use but an approach that could have wide applications.

Techniques such as deep learning could help companies make smarter inferences about their customers, says Matt McIlwain, a partner at Madrona, a venture capital firm in Seattle. They will be able to identify preferences and make predictions, he adds, such as when customers are most likely to want to be contacted or which ones are most at risk of not renewing a contract.

Start-ups flocking to the field face some daunting competition. The biggest advances in AI are being made inside big tech groups such as Google, IBM and Facebook, which have invested heavily in the field. These companies are secretive about exactly how much they are committing to the technology, but have come out with public demonstrations that experts say show they are ahead: a Google test that identified cats from YouTube videos, a Facebook system called Deep Face that recognises pictures of people, and IBM’s question-answering system, Watson.

Entrepreneurs such as Mr Tuttle, however, are counting less on being on the leading edge of a new technology and more on packaging existing technologies into narrowly targeted applications. In the case of Expect Labs, this means a voice-activated service that companies can use to help their customers do things like search through their online catalogues.

Big groups “are trying to build this technology that will solve everything — we’re trying to solve a different problem”, he says.

The basic uses for the technology fall into several different areas. Thanks to improved pattern recognition capabilities, identifying images — a notoriously hard problem for computers — has become far easier. Vicarious, one of the most ambitious companies in this area, recently raised $72m, after demonstrating that it can solve Captchas, the visual puzzles that are used by websites to distinguish humans from computers.

The same technology is also being used to help computers “understand” language — a problem known as natural language recognition. That is one of the techniques behind systems such as IBM’s Watson, which queries large bodies of information to arrive at a most likely answer. A third popular use rests on trying to identify relevance — whether that means personalising online content and other recommendations, or more effectively targeting advertising.

As often with a promising new idea, some of the first applications have been in the financial markets, although the amount of money at stake makes those involved wary of talking.

“If your financial application works, why disclose it and arbitrage it away?” says Babak Hodjat, ‎chief scientist of Sentient Technologies. His company draws massive computing power from data centres to run full-blown simulations of financial markets: by applying “evolutionary algorithms” that try to learn from how markets react in different circumstances, it hopes to develop models for predicting how they will behave in future.

Turning ideas such as these into practical applications in a wide range of fields accounts for much of the investment that is now pouring into AI. Sentient, for example, recently raised more than $100m to apply its technology in more areas, reflecting the higher costs of things like hiring the industry experts needed to “train” AI systems to work in different fields.

The most attractive industries are those with large volumes of data to crunch and high-value problems to solve, such as healthcare, insurance and ecommerce, Sentient says. Computer security and fraud detection are also high on the list of many AI companies.

There are other costs in making the technology functional in real-world applications, says Mr Purpura at Context Relevant: “The real battle isn’t being fought over the underlying machine learning technology, it’s in building support systems to make it usable.” These ancillary technologies include the data “pipes” needed to funnel large amounts of information, he says, as well as control systems needed to make sure AI operates within acceptable business parameters.

With many start-ups under pressure to show that their technologies are more than just impressive parlour tricks, the scramble for investment could determine who will survive the inevitable AI shake-out to follow.

Decoding the language of progress

Artificial intelligence, machine learning, deep learning, neural networks: building machines that tackle problems that were previously believed to be solvable only by the human brain has given rise to a range of techniques and jargon, writes Richard Waters.

As in other branches of technology, disagreements over the best approach can sometimes sound like religious schisms. “The moniker you use describes the tribe you’re from,” says Mr Purpura.

AI, the loose term applied to the field since it was founded, carries with it the dream of full human-like computer “thought”. But attempts to encode human thinking in computer logic fell well short, accounting for the disillusionment that beset AI until recently.

The revival of interest owes much to machine learning, an approach that deliberately steers clear of analogies with human thinking. Machine learning is a product of the collapsing costs of information processing and the masses of digital data that can now be gathered and channelled online. Probabilistic techniques are used to “train” machines as they churn through the data, until they are able to see patterns and reach conclusions that were not programmed in at the outset.

A subset of machine learning, known as deep learning, accounts for much of the latest excitement. Deep learning draws on another idea from the history of AI: neural networks, or software that seeks to emulate the processes of the human brain to “learn” more quickly.

Advances in neuroscience have contributed new ideas to this form of bio-mimicry, says Jana Eggers, president of Nara Logics. The aim, she adds, is to “see how human brains decide on things and make computers do that better”.

Back to the top of the page

Copyright The Financial Times Limited 2023. All rights reserved.
Reuse this content (opens in new window) CommentsJump to comments section

Follow the topics in this article