The 5 A’s of AI

Summary — Fundamentally, there are five ways to work with Artificial Intelligence (AI). The 5 A’s of AI are the five ways for us to choose how we want to work with AI: are we guiding the AI or do we trust the AI enough to let it guide us.

Artificial Intelligence (AI) will have a big impact on our work and life. Just like a simple pocket calculator is better at making calculations, AI is better at specific more complex tasks than we are. Maybe we could do it, but we rather use the calculator — or soon the AI. The difference between a calculator and AI is, that AI can deal with way more complex tasks; even so much more, that in some situations we may choose it to lead us instead of us leading the AI.

Although AI and human intelligence partly overlap, they are very different.

AI is different from human intelligence

The stupid thing to do would be to compete with AI in fields that AI is really good at. And there are many. But there is good news. Although AI and human intelligence partly overlap in function, they are very different beasts. There are lots of things that AI cannot do, that we are very good at. AI is better at well-structured, predictable and repetitive tasks; humans are better at eg. ill-structured, judgmental or improbable tasks:

  • Ill-structured problems — Today, AI is applied successfully to solve ‘narrow’, well-structured problems, such as beating the world champion of Go, with eg. a clear criterium for success. Humans excel in solving general, ill-structured problems, such as answering the question how we should measure the quality of life.
  • Understanding context — AI does not understand context. Even though it may be better than doctors at recognizing skin cancer, AI does not have the faintest notion of what a skin is or what cancer is. To the AI, the skin image is just a pattern of ones and zeros. (A different, even more basic example is the Winograd Schema Challenge.) It is very difficult for AI to give contextual or ethical judgements.
  • Highly improbable events — By the very nature of how AI works — pattern recognition and prediction — it cannot anticipate black swans, highly improbable events. Humans are not brilliant at it either, but better than AI. (It must be said, though, that machine, animal and human behavior is pretty much based on patterns — which leaves a lot of room for AI, still.)

Double Smart: human intelligence augmented with AI

It is clear that smart people who use AI as a tool will be more effective: they will be Double Smart. So, the question is not whether we will work and live with AI, but how we will use it.

So, how will we work with AI?

The keyword in working with AI is ‘augmentation’: acting in a smarter way together with AI. Now, again, the question is: how exactly will we work with AI?

Fundamentally, there are five ways how we can work with AI. I call these the 5 A’s of AI. The logic of the five A’s is simple. At the first A, the AI is a follower. Along the A’s the AI becomes more and more self-reliant and guiding — up to full autonomy at the fifth A.

The 5 A’s of AI are five ways for us to choose how we want to work with AI

But it will be us, humans, who are in control. Even with full autonomy, we are the ones who interfere or switch the AI on and off. AI is a tool, a smart tool, but still a tool. Nothing more, nothing less. The 5 A’s of AI are five ways for us to choose how we want to work with AI: are we guiding the AI or do we allow the AI to guide us.

So what are the 5 A’s of AI?

(Absence)

Not so fast! There is a sort of ‘zero-th’ A. For the sake of completeness. We already use a lot of tools where AI seems to be absent, but isn’t. It’s there often without us knowing. Eg. face recognition in your photo app, the spam filter in your email, personalized online shopping recommendations, or fraud detection for credit cards. Sometimes you wonder “how on earth did [my machine] know that …?” That could well be AI ‘in stealth mode’.

So, again, what are the real 5 A’s of AI?

1 — Assistant

You use AI as a servant, answering your questions and executing your orders. The AI basically is an Expert System, with mostly predefined functionality and increasingly one that you can speak to in natural language (NLP). It may have limited predictive functionality: the ability to anticipate what will happen (inherently probabilistic). The obvious example here is telling your smartphone what to do or asking for information. “Alexa, what’s the weather, today?” or “Hey Siri, invite Julia for dinner tonight 8PM in Le Paris Restaurant and book a table for two.” No need to say ‘please’ or ‘thank you’ to Alexa or Siri. The Assistant AI may ask questions, but does so only for clarification.

2 — Advisor

You use AI as a buddy or consultant, giving you tips and suggestions with alternatives. (The start of) a real two-way dialogue. This is the enhanced version of an Assistant AI: next to significant predictive capabilities, it has prescriptive capability, it is able to give recommendations in order to achieve goals. Google Maps is a simple example. If you ask for directions, it gives (prescribes) you alternatives for the fastest route (the goal) with expected travel times (predicted probabilities), eg. avoiding traffic jams or toll roads. A better example is a smart AI expert system for psychiatrists. In psychiatry, often there is not one clear generic answer. It depends on the patient. Rather, the psychiatrist gets cure suggestions with healing probabilities — which in a man-machine dialogue the psychiatrist then can translate into a custom solution for her specific patient. The Advisor AI is different from the Assistant AI in that, when the AI communicates, it brings new insights and suggestions into the dialogue, rather than just ask for clarificiation or execute orders.

3 — Augmented

You do what you are best at, AI does what it is better at than you. Together with AI you do new things that you could not do on your own. You achieve more than before, but now you only do the fun part. AI does the predictable and repetitive work — no matter how complex — in a semi-autonomous way: it does it’s subtask, but does not complete the entire task; a human does the other subtask. Or we as humans pilot the AI, tell it what to do, control it, have the overview or complement it. An example is translating a text eg. from English to German: AI does the initial translation, and then a human translator corrects the language. Similarly, AI is used in law firms to search through millions of pages of case law and write a draft document. An experienced lawyer then completes the document. A different example is working with smart robots in production work, where the smart robot does the heavy and boring work fast and accurately.

4 — Authority

You follow insights and instructions from AI, because you trust that the AI knows best. Sometimes it even tells you what to do — as a planner or supervisor … or boss? The Authority AI is fully descriptive, predictive and prescriptive. It ‘knows’ better than the humans who work with the Authority AI, because it has access to and ‘overview’ of more and better data. In a gradual process of working with an Authority AI, human collaborators have step-by-step learned to trust the Authority AI, because agian and again it turned out to ‘know’ better, more effective or efficient ways of working than humans. An example are the smart mobile apps, that give parcel delivery persons or pizza delivery boys instructions on when, where and via what — dynamically optimized — route to deliver the next parcel or pizza, including evaluation of delivery performance and rewards.

5 — Autonomy

You start and stop the AI, performing the task at hand on its own. Self-driving cars are the obvious example. You step in, tell the car where to go and sit and relax. If things go well, you do nothing. With the right authorization and tools, you can interfere with the AI at any time. Autonomous AI goes a step further than autopilot systems, currently in use in eg. trains and airplanes, in that it dynamically learns and improves its performance — it is not rigid, pre-programmed instruction based software code.

In the course of time, many autonomous AI systems will feel so natural, that they will appear to be absent. We have then become so used to them, that we do not consciously observe and notice them anymore as AI systems. That will close the loop of the 5 A’s. (See image below)

Is there one more A? Maybe.

(Adversary)

There are lots of spectacular, imaginative stories of robots and AI outsmarting humans, taking over the world and ruling the universe. So far, I have no signs of that, but I keep an open mind. To be honest, I’m much more worried about unethical use of — eg. biased — AI by fellow humans.

I think the opposite is more likely really.

(Affection)

No joke. People do have passionate romances with Siri and Alexa. And people do bond emotionally with humanoid robots — when they fall, we feel the hurt. And smart robots are being used to accompany demented elderly. But what I mean here is AI recognizing and responding to your emotions. It already happens in some customer service chatbots.

>>> In case you missed it: read all about the AI Love Affair.

It may take some time before we build affective relations with AI in large numbers, although we already are pretty much addicted to our smartphones … (yes, another A). In the mean time, I suggest we first get used to working with AI in each of the five described ways, the 5 A’s of AI.

Case Greenfield

. . .

This article was published earlier on LinkedIn: https://www.linkedin.com/pulse/5-ai-kees-groeneveld/

Thank you for reading my post. Here at Medium and at LinkedIn I regularly write about management, work and life with Artificial Intelligence. If you would like to read my future posts then simply join my network here or click ‘Follow’. Also feel free to connect on Twitter.

Follow, Clap or Share if you appreciate this topic.

Note — The 5 A’s of AI are presented under the Creative Commons Attribution-NoDerivatives license (CC BY ND). This means you are free — and kindly invited — to re-distribute and re-use the 5 A’s of AI as long as Case Greenfield is credited as the creator.