The discussion in Chris's thread about Google's latest telephone bot made me want to start a conversation about personal agent software. Decades ago, I remember thinking, wouldn't it be nice if I had a program that could do stuff for me while looking out for my best interests? It would know my tastes and also know how to get the best airfares, make restaurant reservations, manage my finances, etc. The technology then was nowhere near ready, but now it's getting close. But the bigger picture is full of dragons.
I suppose the question comes down to this: is there a viable business model that could support a true personal agent, one that puts my interests above corporate shenanigans? I would be willing to pay for it, but It would need to be highly secure and trustworthy. I'm guessing that it would need to come from some sort of open-source based organization. But I have no idea whether this is even a viable concept in the age of the Apple/Amazon/Facebook/Google overlords.
I definitely understand the desire not to give big corporations even more control over our lives and information, but I think it's pretty unlikely that open source organizations or individuals will have the ability to develop or provide access to good AI agents on their own, at least not within the next couple of decades.
The reason we're only seeing big corporate players like Google, Amazon, Apple, and Microsoft do useful things with AI voice bots right now is that it's really hard.
Not only does it require millions of dollars worth of R&D (that's basically the salary and expenses of many smart people working for many years) to develop the necessary software and train the AI models, it also requires significant hardware resources in order to provide the compute power necessary to run these advanced AI models in real time for millions of users around the world.
Over time we'll see personal devices like phones become more powerful and more capable of doing local AI processing, but it'll be a while before your phone is powerful enough to handle all the computation for the kinds of things Google Duplex is doing. And even once that happens, the software will still be hard.
I could maybe see an open source-friendly company like Mozilla (which operates as a combo of a nonprofit organization and a corporation) eventually making small inroads in AI, but it's hard to imagine them being able to compete on quality with the big guys. I just don't think they have the money that'd be necessary to fund the work.
I think the best we can hope for in the near term is for more companies to do things the way Apple does. Apple's a big profitable corporation so they're able to invest the millions of dollars and years of effort necessary to make major advancements in AI, but they also have a very good track record of caring deeply about user privacy and building features in such a way that privacy is protected (to the extent that it can be).
What I have in mind wouldn't replace corporate agents, but rather work on my behalf within the existing Internet ecosystem. A meta-agent, perhaps. It could get all the cycles and storage it needs from the cloud. It wouldn't have to reinvent the wheel. There are already a number of speech to text APIs that small apps use as well as an increasing number of machine learning frameworks in the public domain. It's reasonable to expect that over time, many more tools and services will become available.
The problem space is smaller--it wouldn't have to understand the whole world's preferences, but only mine. It might not need to learn from billions of data points. In fact, one-shot learning (or a few) is an active area of AI research. Humans can and do learn from very few examples. When it doubt, it could ask me--it would be in my interest to train it. It would be up to me to control what the agent knows about me, with the understanding that I might be limiting its capabilities if I choose to limit its access too much. I might also specifically instruct it to avoid certain businesses or services that I find personally objectionable.
This wouldn't solve the problem of surveillance capitalism--my agent would still be dealing with businesses that are sucking up as much information as possible from every interaction. But if done cleverly, it might mitigate the problem a bit. At the same time, it might make it easier for me to find the best deals on whatever by checking more sources than I would ever bother with myself or relying on consolidation sites that may have their own agenda . It could also alert me to news and events based on my preferences while keeping my overall profile hidden.
But the question remains, is there really a market for this? Or have we just all been assimilated?
It will happen. Just a matter of time - hopefully not too much time. Evolving and learning AI writing programs will create these but that's on a longer time frame. Not sure how it'll develop in the short term though I think it's a technology that can be implemented gradually to do more and more.