We're going to be the heroku for autonomous intelligent agents. If you're interested in distributed cognition, cybernetics, AI, cloud robotics, probabilistic robotics, multiagent systems, ambient computing, hit me up. I'm @dpg on Twitter or email me dan.p.gailey@gmail dot com
I basically sold everything I owned to crash in my van in Palo Alto and build this.
Is there a market for traditional AI techniques as a service? Or maybe the wording is just throwing me off? The word 'agents' in particular is used a lot in 60's AI papers (Minsky's influence), so maybe I'm misinterpreting?
If modern machine learning models are better for every application, then that's most likely what we'll support. I think symbolic AI still holds some merit when thinking about how to interact with traditional physical interfaces, no? There seems to be a complement in the methods when considering a full stack thinking and acting approach, but perhaps I'm being naive.
For example, if I'm a robot in the world, it would be easy for me to access interface knowledge, and how I could interact with it to complete my task, without needing to reason what the interface does or is. This gives me a basic understanding or knowledge base. Now if there is a higher level method of thinking that produces ingenuity, and as a result a more efficient method for solving a known problem, then I assert more modern techniques can be used to adjust older symbolic representations.
I'm just imagining how an android/autonomous agent would need to operate in the real world, perhaps akin to how we might perceive and function within it, a posse ad esse.
Also I use 'agents' because I am going back through older research and it seems to fit the description of the entities involved (software, hardware, robots, devices, people). I don't see any reason for some of the vocabulary or terms to have fallen out of favor.
In a world of autonomous intelligent multi-agent systems, there is a market for knowledge representation about the world, how to complete tasks, as well as higher level thinking/reasoning functions, et cetera.
This does indeed not preclude new methods. SLAM is of course what everybody uses in autonomous robotics. But you also have reinforcement learning and imitation learning and a wide diversity of representations and models from DEC-POMDPs to Bayesian models to neural nets. Each time you'll need somehow to decompose into modules, units, etc. Agents represent something in the real world and are often the right term for a module that has (virtual) sensors and actuators.
Apart from that I would recommend to focus on something, so you can really excel in that part of AI.
We have chosen "understanding where an AI is located in the real world" as our main goal. I can imagine there is something similar you would like to achieve for AI.
Just my two cents! Good luck!!!
The platform I'm currently building is just to facilitate the provisioning and correlation of third party services for emergent results.
Thanks again for the info. I'll conduct more research. Cheers. :)