When talking about AI, individuals not versed in practical contemporary AI generally assume the subject of reference is True AI - that is, AI which is cognitively equal (or greater) to a human, and generally behaves in a similar way. The kind of AI that is depicted in cinema and video games - AI that’s fully emerged, possesses general intelligence, and is capable of interacting with other human beings in an organic way (be it peacefully or violently).
I have discovered this while giving a talk at the Flick The World tech festival in Zurich, when people began telling me that as far as they are aware - AI does not exist, or suggesting that the term Artificial Intelligence ought be changed to Artificial Stupidity.
The people in question were not disconnected from the tech world - on the contrary, they are involved deeply in the world of tech; one was the very organizer of the festival itself, a person in charge of their own Hacker-space and offering open source tech education to others, another was an individual working on microprocessors within the renewable energy industry, and yet another running an interesting initiative of personal servers and portable networks.
What more, at the very festival itself, there’d been a little project with physical microprocessors hooked up together in a chain, mimicking networks of neurons and producing spiking electronic pulses translated into sounds.
This disconnect between a tech which was attempting mimicry of neural nets and the simultaneous mislabeling of this as non-AI related, and the disconnect between the individuals involved in it and their definition of what AI is, had really made me think, and forced me to re-frame AI within new context for myself.
Prior to this point in time, due to my own social circle and background, I was under the wrong impression that everybody knows what AI means, or at the very least - I simply hadn’t thought of it, and took my own knowledge for granted.
But thanks to these questions that have been asked of me, I learned that the general definition of AI is mainly reliant on pop culture’s depictions of it - depictions of a True AI, AI that’s interactive, broad, and various in function, and not the formal definition of an “Agent that is capable of perceiving it’s environment and taking actions within it to maximize it’s chances of achieving it’s objectives”.
I used this knowledge to express the formal definition in pop culture related terms within my class, but still this has not left my mind.
Namely, the notion that it is not the individuals which ought re-learn the definition and re-adjust themselves to what AI actually is, but that it is AI’s job (or rather, it’s makers’) to learn to broaden the function of AI and bring it closer to the popular idea of what it should be.
I’ve spent a long time fantasizing about all sorts of interactive AI visions; AI at bus stops or at the corners of a building, AI as vendors at the store, AI as information desks in stations and in airports, even AI as advertisements or as moving posters promoting relevant events. Yet after these thoughts of “definition” and “broad knowledge” were planted in my head, the embryo of a new project slowly began gestating within my depths of thought - the embryo of interactive, varied AI with a tangible presence, which would fulfill the general understanding of what AI is better than what AI generally does today.
Not to contrast the fact that nearly all AI is interactive, but to address the level of interactivity itself, I have set out to build a Virtual Intelligence Terminal - a multi modular AI with presence in the world much more akin to that of yet another person, than the types of interactive AIs that are currently and (relatively) broadly known.
See, AI is very often interactive in our day to day use of it within our various websites and applications, the level of interactivity usually ranging from the AI suggesting to tag familiar faces in your photos, to fully conversing with you under the guise of customer service in company website chats. But looking at all these interactions, they do not pass as AI within the constraints of it’s pop culture definition - these types of AI; practical and purpose driven, are often quite one-note and monotone. This very purpose driven practicality, most often serving a business interest of some sort, are loathe to truly classify as AI within pop cultural constraints. Rather, they are more similar to robots, no matter how clever and dynamic their systems are. They serve one purpose, perform one duty on the so called "production line" of business, and are not much good for anything else.
A very specialized AI that’s capable of doing one task well but cannot do anything other, is not AI at all - it is a task completer. A single purpose, one use tool, just like the robot that tightens bolts on a production line. Even if the robot’s “mind” is much more flexible in the performance of it’s task - the fact that it can only still be used for that one task and that one task alone, makes it, in some ways, undeserving of the title of AI within the eyes of the broad public.
There are, of course, more exploratory and experimental AI interactions out there - interactions such as the image transformation offered by the (relatively) well known demonstration of the Pix2Pix paper, or chat bots with broader conversation capabilities like the relatively old and well established Cleverbot, or Microsoft's Twitter bot Tay. Yet all those, too, offer merely one type of interaction, usually driven by research goals, and even Microsoft’s Tay - the Twitter bot AI - has ro”bot” in it’s definition’s title.
All such AI, unless incorporated into products which offer a broader function, rarely see much use and interest beyond their novelty expiration date. Even those incorporated into well used services and products, like Facebook’s facial recognition AI, are not addressed on their own, rather they are thought of as but a minuscule component of a larger entity - that is, if they are thought about at all.
So then, what if there were an AI that’s varied in it’s interactivity?
AI that has a multi-functional core, is capable of doing varied things with varied people, and takes engagement one step further by being represented via a physical terminal?
An AI that’s isn’t driven by a business or research goal, but is instead driven to fulfill the pop culture definition of AI - it’s main goal to appeal to the expectations of others, rather than to achieve any specific task?
Instead of trying to answer these questions theoretically, I have set out to build a Virtual Intelligence Terminal that tries to accomplish just that.
The goal in mind is not to reinvent the wheel, but rather to use existing knowledge and technology in order to assemble a Frankenstein of sorts; a virtual intelligence that’s there just for the sake of being itself, under constraints of general understanding of what AI should be.
The goal is thus as such:
To make an interactive physical terminal that’s capable of visual expression, audible speech, voice and facial recognition, which has a limited capacity for memory of people and of it’s disposition towards them, and which can make requests, provide rewards, play games, give information, and simply chat with people.
Not to delude myself in expectation that such a project will be easy, even if using available existing tech, I fully recognize that I don’t know exactly what I’m treading into - but also I have no intentions of finishing the project until all the objectives are complete.
In the next series of posts which will be titled “Building a Virtual Intelligence Terminal” I will be documenting every step taken towards this goal; every decision made, the reasoning behind each choice, and the accompanying tutorial to reproduce the same exact results and make your very own V.I.T. as you desire.
As a final footnote - I use the term Virtual Intelligence instead of Artificial Intelligence within this context so as not to create the expectation of myself that I am building a general AI, but rather yet another bot, one that is simply capable of mimicking a real AI a little better, and one which can appeal to broader definition of AI more readily than other practical AI that’s out there.
Comentarios