MIT EmTech Conference
I spent the last couple of days in Boston at the MIT Technology Review’s EmTech conference and some of my immediate takeaways are:
Artificial Intelligence might be a catchall phrase for machine learning, linguistic programing, advances in one shot learning and automated interpretation of optical data among others but all these strands are experiencing enhanced growth. The field of artificial intelligence has been gestating for decades but the evolution of large data sets gives many of the theoretical applications that have been confined to universities room to grow and reach commercial utility.
The ability of computers to read text and understand what has been written has improved, meaning they can read invoices and check them against contractual agreements to streamline middle and back offices for a host of large corporations. The majority of companies offering these kinds of service are either privately held or represent a small part of the company’s overall business. The impact on the companies purchasing these kinds of services is likely to be highly beneficial and could represent enhanced productivity which could boost share prices. .
Autonomous vehicles are another theme where expectations are high and a competition is evident in which company can make the boldest claim. NextEV, a Chinese start-up, plans to have its electric supercar on the roads of China by early 2017 and has its sights squarely set on Tesla’s pre-eminence. The CEO of Nutonomy, Karl Iagnemma, which has a fleet of autonomous vehicles on the streets of Singapore offered a more measured and altogether more grounded approach.
His point was that the technology is developing rapidly but it will not be a simple matter of flicking a switch and the world will suddenly move over to autonomous vehicles. He said that where he expects the greatest near-term utility for his cars is in the last mile of a commute. Often the end of the line for public transport means that commuters have difficulty getting a taxi to cover the last mile in less populated areas. This would be the perfect environment for autonomous vehicles to prove their utility because they would face little competition and traffic is quiet.
I was a little disappointed that the cutting edge of robotics was not discussed more fully. The latest video from Boston Dynamics, focusing on its Spot Mini, released in June, is orders of magnitude more impressive than what we saw from the company at the beginning of the year. It highlights just how much progress is being made but also how much farther ahead some teams are than others in developing the technology.
Stephanie Tellex from Brown University gave an inspired presentation on her project to use the more than 400 Baxter robots in research facilities all over the world to conduct machine learning exercises during down time. Her clearest point was that most machine learning now takes place using photos taken by humans but it is much easier to teach robots using the type of first person experience that a child has when turning a switch on and off or constantly opening doors. That is why she is programming her robots to film their actions from multiple different angles and through trial and error resolve the question of how best to pick an object up.
I’ve previously written about how teaching robots to have a tactile experience is intimately linked to optical data. The way researchers are now focusing on allowing robots to gather optical data on the exact task they are set to complete is a recognition of that and echoes the process Softwear employs to develop sewing robots.
It’s going to take time to commercialise but the presentation given by Shyam Gollakota for Jeeva Wireless was perhaps the most exciting thing I’ve seen in a long time. You might remember that an article from the Daily Telegraph a year ago highlighted the development of a company called Freevolt which harvests energy from unused ratio waves but I have not seen anything since about the company’s plans. Jeeva does the same by picking up ambient Wi-Fi and radio signals to power microprocessors and transmitters so that sensors embedded in everything from clothing to contact lens will not require batteries. This video highlighting this interscatter approach from the University of Washington may also be of interest. The relative cost of producing these kinds of products has the very real capability of ushering in an era both of embedded technology but also disposable technology.