The ISAs. They Are Not Here Yet.
Writing, the printing press, industrialisation, the microchip, the internet.
There are several inventions that have sliced history at the point of creation, that changed the world so much that you can discuss civilisation has ‘before this’ and ‘after this’.
I expect ISA, Intelligent Software Agents, to be the next one in that list.
What is an ISA? For those who don’t know, an ISA is like a software butler. Not a robot you order around the house but an ethereal ghost in the machine that can follow general instructions and perform tasks for you. Basically, take a general AI and give it a couple of white gloves and a double tail jacket.
And while some of you might be thinking we have Siri and S Voice, the Google Assistant and Alexa. But these are an OK start. We are not at the goal post yet. While I might be able to tell a properly equipped smart home to set my AC to certain temperature or start my coffee maker, I can’t tell a digital assistant to sort my news.
This is my benchmark now. Can we have a software agent that sorts my news?
What do I mean: I have a news reader on my phone that scrapes certain feeds I set it too. Sometimes I read my news while doing something else, sometimes I want to focus on that, sometimes I listen to it as I walk the dog and have distractions and sometimes I listen to it on my bicycle and want something more intense.
Currently I sort it by feeds and categories which is the only way I can. And I try to number categories by rate of intensity. So I read it in ascending order and listen in descending order. But if I had a proper ISA, I could tell it to put the short Slashdot articles, the ‘list of things you should know’ articles, the articles I might be interested in the title of or the first couple of paragraphs, in the light category.  I would tell it to put the detailed, statistics heavy articles with lots of reference images in the ‘things I really need to sit down and read’ category. I would tell it to put the long analyses in the ‘listen when out and about’ category. And I would tell it to put APOD and Digital Blasphemy and articles that are just video references in the ‘check this when on a PC’ category.
But we can’t do that yet. When we could, that would be the shit!
Posted in IT, Practice, Thinking Out Loud by Eran with 6 comments.
I don’t think there will be a point to witch you’ll point and say “there – there is where ISA started”. It’s much more likely that the existing agents will get more and more optimized, better, smarter if you will, until one day you’ll discover we have ISA’s and realize we’ve had them for a while.
I think in this case, there are two options. Either there will be what you said where AIs will be slowly taught to do more tasks and more tasks until they can do quite a lot.
But it is also possible that there will be a breakthrough and someone might just invent some kind of algorithm or system or training technique that will facilitate a general purpose AI. In that case, there will just be an announcement of the idea and quickly the AI will be taught to do all the things.
I think your use case sounds like an edge case for “AI helpers” as we know them now and we’ll likely not be resolved before we have a real “general purpose AI” that can reason about human language – which it could be that such a thing is impossible, or just so amazingly expensive that it will never be available for the general populace.
That being said, your use case sounds like something that can be solved rather easily by the current state of the art machine learning technology – you just need to have some experience in programming and to play a bit with the tools (you don’t even need to know any math to do it, today’s tools are very easy to use). There down side is that it will require several weeks if your time to figure out the tools and run the training, and the run time will probably cost a bit more than you may be willing to pay for such a service.
That’s what I mean. It’s a benchmark for general intelligence. And I don’t really have the time or money to train an AI myself.
Then the question remains, as Ayal noted, as money and personal investment requirements for building a machine learning based classification system (note that I take pains to not call this AI) keep going down all the time, will there come a time that you will be willing to make that investment in order to achieve your custom use case, without the need for a general purpose AI? Most definitely!
Will that time comes while you are doing other things and then some time later you are explaining your ISA wish list to someone and they go: “oh, sure you can do that right now, here – let’s log into AWS Alfred and set it up”? Maaayyybbbbeee….
Maybe. But, more likely, I will think of another benchmark and another and maybe we can teach them all and it’s definitely moving but it’s also definitely a target many are aiming for.