Sentient Technologies

» see sentient.ai as it lived, on the wayback machine

~ the last words ~

The company was founded in 2007 under a name nobody remembers, Genetic Finance Holding Ltd, and spent seven years in stealth. Two of its three founders had the best possible credential for an AI startup in that decade: Babak Hodjat and Adam Cheyer worked on the natural language technology that turned into Siri, which Apple bought and put in every iPhone. Antoine Blondeau ran the business. The plan was to apply evolutionary computation, which is to say algorithms that breed and cull candidate solutions over generations, to problems with money attached.

It emerged in November 2014 with $103.5 million in Series C funding, described in the coverage as the biggest single round any artificial intelligence company had taken. Horizons Ventures had led an earlier one. By 2016 the total was over $143 million and Sentient held the title of most heavily funded AI company in the world, a title that reads as comedy now given what the same label costs today.

The technical claim was real and genuinely strange. Sentient ran its algorithms across something like two million CPU cores and five thousand GPUs spread over 4,000 physical sites, assembled from spare capacity rather than owned data centers. Few distributed systems of that size had ever been pointed at machine learning. It worked with MIT's Computer Science and Artificial Intelligence Laboratory on predicting sepsis in intensive care patients from blood pressure data.

The commercial products were much smaller than the machine. A visual recommendation engine for online retail launched in 2015 with the Vancouver shoe store Shoes.com as its first customer, and the company ran a hedge fund of its own, Sentient Investment Management, on the same technology. The product that lasted was Sentient Ascend, launched in September 2016, which used the evolutionary methods to optimize web pages by testing many variants against each other at once instead of two at a time.

~ the record ~

~ a hundred and forty three million dollars ~

Money was the famous part. Sentient turned up in the funding tables and on the conference panels about where the industry was heading, and it never turned up in anyone's daily use, because nothing it built was a thing a person could open. Shop at Shoes.com in 2016 and you were using Sentient without a single clue that you were, which was the entire design rather than an accident of marketing.

Enterprise AI companies of that generation lived or died on whether the research could be packaged into something a mid-sized company would buy for a recurring fee. Sentient had the research and never settled on the package. By the time it needed to, the foundation model wave was starting and the money that had made it the best-funded company in the field was looking at very different companies.

~ what the algorithms actually did ~

Evolutionary computation is older than deep learning and was out of fashion for most of the period when deep learning was eating everything. You define a fitness function, generate a population of candidate solutions, score them, keep the good ones, mutate and recombine, repeat for thousands of generations. It is slow and it is embarrassingly parallel, which is why a company with two million borrowed cores would choose it.

What Sentient added was combining this with neural networks, using evolution to search for network architectures and strategies rather than hand-designing them. The approach has a name now, neural architecture search, and it is standard. In 2016 it was an unusual bet made by a company most engineers had never heard of.

~ what came after ~

The breakup was quiet and quick. Cognizant, the IT services giant, bought a set of assets and the research team in December 2018 and talked about it publicly in February 2019, rebuilding the work as an "evolutionary AI" practice inside a consultancy. Sentient Ascend went to Evolv Technologies in March 2019, a company formed specifically to take it, which raised $10 million and carried on selling the optimization product under its own name. The investment arm was sold separately.

Sentient Technologies itself was dissolved that year. Both the research and the people carrying it came through intact under other logos, so what the hundred and forty three million dollars finally bought was an acquihire with extra steps. Babak Hodjat became Cognizant's chief technology officer for AI and is still there.

@}-->--

              )
             (_)
            .-'-.
            |   |
            |   |
            |   |
            |   |
          __|   |__   .-.
       .-'  |   |  `-:   :
      :     `---'     :-'
       `-._       _.-'
           '""""""

« previous  ·  random  ·  next »