GUEST:
After a recent seminar at Stanford, I had a chat with adjunct professor Jerry Kaplan about artificial intelligence embodiment. The question was: Who are the thinkers and companies who are really pushing AI theory? Some experts believe that for artificial intelligence or artificial general intelligence (AGI) to function peacefully and effectively in society, it needs a body. Thoughts vary on the level of embodiment, the mortality of that body, and the complexity of sensing abilities and empathy needed.
So I did some research into the current state of robots and AI embodiment to identify the clusters and trends in this area of technology. Below is a short summary of what I found. It’s not complete, so please feel free to tweet me with suggested updates.
Using Quid, a research tool that visualizes company data and clusters through natural language processing (NLP) and some fun machine learning (ML) algorithms (disclosure: I used to work at Quid), I looked at companies that use AI in robots or AI embodiment in general.
The leaders in each of these areas may be on the cutting edge of not only AI functionality, but also of AGI theory. Understanding how the fields interact, where interconnectedness is most dense (like between autonomous vehicles and future of work) or where there are gaps (like between process automation and environmental sensing), can help us understand areas for innovation and maybe even new AI theory.
I found that, generally, embodied AI companies cluster into the following areas.
Personal companions/child cluster
Companies like Jibo and Mabu from Catalia Labs provide companionship and care for kids and hospital patients. Because of the NLP, empathetic sense, and domain expertise needed for human interaction, a huge amount of innovation and money coincide in this area. And because a robot can never truly know if their goal of, say, making the patient feel peaceful, is 100 percent satisfied forever, growing AGI in these service settings may be an important way to control for catastrophic AGI takeover.
Operations/movers cluster
These are robots with advanced physical capacities. They have a focus on supply chain or operations, either in robots or enabling software and applications. The popular Roomba — a widely used consumer robot for home cleaning — falls into this cluster. A lot of old-school labs like Charles River Analytics and Boston Dynamics work in this area, with their associated datasets, institutional knowledge, and existing theories.
Environmental sensing systems cluster
These companies focus on interacting with the natural environment and involve sensing and feedback between intelligent systems and the real world. Honda works in this space developing systems for environmental sensing and adaption. Also fun is Gridbots, which makes robots for underwater military and industrial use.
Mavrx is another example that uses high-resolution crop imagery to create connected and intelligent systems for agriculture. This particular company makes me wonder if the goals of achieving sense, balance, complex feedback, and flexible reaction to extraneous events are “fertile ground” for AI theory.
Legal AI cluster
This cluster includes AI companies focused on the legal system or applications for document management and analysis. At first glance, this does not seem to really push the concepts of AI and is more about practical use of machine learning, but feel free to set me straight. Clearly, there is a lot of money here, and where money and data exist, AI may flourish.
Virtual reality and future of work cluster
This cluster includes gaming software innovator Velan Studios, which provides software for the integration of VR and robotics. It also includes a number of AR, VR, and “mixed reality” companies using advanced ML to deliver their solutions, often for games. For example, Sony PlayStation and other gaming labs are currently looking at empathetic avatars.
Will any of these companies ever get enough data to create AGI? Maybe not, but the novel combination of sensory capabilities, empathetic systems, and feedback is rich ground for AI theory.
AI software developers cluster
This cluster of companies is key to the fields of AI and robotics. Such companies develop solutions for AI, speech, and industrial processes. Included in this space are Pony AI, The Curious AI Company, Osaro Inc., and my personal favorite name, Twenty Billion Neurons GmbH. Given time, I’d like to dive more into the promises and insights unique to each actor in this group.
Process automation and consulting cluster
This robot-centric group of companies either develops in-house or consults on development products for industrial, financial, and manufacturing process automation using AI, but most often low-intelligence robots. As researchers develop sophisticated robots, AI theorists may need to look to this group to understand practical feedback mechanisms for AI development.
Autonomous vehicles cluster
In this cluster, we see companies, investors, and funds newly set up to specialize in robotics and AI. For example, Robik is looking at last-mile delivery with intelligent robots, and Toyota is developing and investing in autonomous vehicles. The ideas these companies employ around dealing with ambiguity could offer good input for AGI theory.
China/manufacturing cluster
A sub-cluster of industrial AI and robotics focuses on manufacturing tech developed in China and Russia, mostly in Shenzhen and Guangdong Province. Shanghai Huoshanshi is the most active investor, followed by Warburg Pincus and Banyan Capital.
Natural language and chat tech cluster
This cluster, as you’d expect from the name, is all about interactive dialogue with semi-intelligent AI and robot-embodied AI. If you’ve read much on singularity (e.g. the easily digestible Avogadro Corp), you’d theorize on how AI in this space might be the most data-enabled for advanced sentience. However, it’s the application of this sophisticated technology with embodiment that may inform a new set of theories for growing AI.
Security and rescue drones cluster
With innovation focused mainly in San Francisco, this cluster of aerial drone companies used for security, rescue, and surveillance is small but growing. Companies include Neural Robotics, Iris Automation, and Aeroxo. The computer vision needed for this field may be an interesting addition to what we should assume robots can sense and react to.
Summary
Here is a heat-map of each cluster, which shows timeline, funding, and number of company sums, as well as some summary statistics about the network.
The top cities for this area of technology are Beijing, San Francisco, London, New York, and Tokyo, in that order. The top funders, again in order, are Intel Capital, Andreessen Horowitz, GGV Capital, Samsung Ventures, Banyan Capital, and Fenox Venture Capital. The number of companies has grown exponentially since 2013.
Many thanks to Kate Montgomery for her help proofing, to Jerry Kaplan for the idea, and to Mark Sagar for the illuminating chat on embodiment theory in Auckland last year.
*My boolean search term was ( ’embodiment’ AND [AI](AI OR “artificial intelligence”) ) OR ( embodiment AND robot * ) OR ( “artificial intelligence” AND robot * )
This article was originally published on Medium. Copyright 2018.
Bethanie Maples is a cognitive development AI researcher at Stanford University and an adviser at Mappr, Aera Technology, and Predicta.
No comments:
Post a Comment