Can AI solve your problem?. Three simple heuristics for recognising… | by Daniel Bakkelund | Nov, 2023

In a product organisation aiming to assemble AI capabilities into their providers and merchandise, there could also be always the issue of bringing the non-AI-literates onboard the AI observe. Whereas not everybody have to be an AI educated, it’s important to have as many as potential contributing with ideas and prospects of exploiting the ability of AI to propel the company to the following stage. That is relevant particularly to space consultants and product people, who’re on excessive of the problems their providers and merchandise attempt to resolve, and realizing the place the shoe pinches.

One downside I’ve realized is prevalent, is the important question of “Which points can we treatment with AI?”. A question that’s surprisingly laborious to answer when posed by a non-expert. So I’ve devised three heuristic questions that it’s good to use each time you’re taking a look at a difficulty, and also you’re questioning “Can this be solved with AI?”. In case you possibly can reply certain to all three of them, chances are high you’ll find yourself in place to begin out an AI enterprise.

You probably can contemplate an AI as an oracle that options questions. What it’s good to ask your self about, is:

Can you particular, in writing, the question you need answered?

That’s, in spite of everything, a examine that applies to one thing you need to do. In case you want to do one factor, nevertheless you’ll be capable of’t formulate what it’s you want, you most likely don’t truly know what you want. Launching an AI enterprise isn’t any exception to this rule.

Occasion inquiries to ask an AI might very nicely be

• Is there a canine on this picture?
• What’s going to the local weather be tomorrow?
• What are subsequent week’s lottery numbers?

All of these are properly posed questions which may be requested. Nevertheless not all of them might be answered, so we’d like one different examine.

We’re capable of contemplate the oracle as a function mapping inquiries to options:

The circle on the left accommodates the entire questions, and the circle on the right accommodates the entire options. The oracle is the function sending inquiries to options. The following issue to ask oneself is:

Does the function exist?

This may occasionally sometimes seem odd, and it’ll get queerer nonetheless: it’s greatest to ask this question on a metaphysical stage — is there any theoretical danger for this function to exist? Permit us to have some examples:

We’ve all seen AIs answering the “canine throughout the picture” question, so everyone knows that this function exists. We’ve moreover seen the local weather forecast, so everyone knows it’s potential, to some extent, to predict tomorrow’s local weather. Nevertheless there isn’t a technique to foretell subsequent week’s lottery numbers. And the rationale for that’s that the lottery is rigged exactly with the target of this function to not exist. It’s inconceivable. And that’s what I indicate by “on a metaphysical stage”.

Why is that this mandatory? On account of machine learning (which is how we make AIs) is about attempting to approximate options by learning from examples.

If now now we have quite a lot of examples of how the function (i.e. oracle) should behave, we are going to try and be taught this behaviour, and mimic it as intently as potential. Nevertheless you’ll be capable of solely approximate a function that exists.

Admittedly, all of it’s a bit abstract, so I prefer to suggest altering this heuristic with the subsequent meta-heuristic:

Can a well-informed human do the job?

Nonetheless metaphysically, given all of the data on the earth and limitless time, can a human reply the question? Clearly, persons are pretty good at recognising canines in footage. And folks did develop local weather forecasts, and do them too. Nevertheless, we aren’t ready to foretell subsequent week’s lottery numbers.

If in case you’ve got come this far, answering certain twice, you’ve gotten 1) a properly posed question, and a pair of) you already know that, a minimal of in concept, the question might be answered. Nevertheless there could also be one other subject to check off:

This one is a wee bit further technical. The necessary factor to the question is that the oracle function often desires further information than merely the question to go looking out the reply. The educated human being, doing the job as oracle, might need further information to resolve or produce an answer. That’s what I talk about with as a result of the context.

As an illustration, the local weather forecast oracle should know the current meteorological circumstances along with circumstances from some days once more to do forecasting. This information is not going to be contained throughout the phrase “What’s the local weather going to be tomorrow?”
Alternatively, throughout the case of pictures of canines and cats, the context is throughout the picture, and no further context is required.

The reason why that’s mandatory, is that after we observe an AI, the AI is obtainable with questions of the type

The AI then makes a guess sooner than receiving the true reply, and over time it’s hoped that the AI might be taught the excellence between cats and canines. Nevertheless for this to happen, the excellence must be on the market, so that the AI could be taught to determine the excellence. Throughout the case of pictures, that’s simple — you merely need to confirm the photographs are of satisfactory top quality to make the excellence potential. Throughout the case of local weather forecasting, it turns into further refined — you actually ought to make an educated decision to what information is required to make a local weather prediction. This generally is a question best answered by space consultants, so you may have to attain out to get an excellent reply to this one.

Nevertheless the bottom line is: if there could also be not adequate information on the market for the educated human to answer the question, then there could also be little hope for the AI to find methods to answer the question moreover. You need that context.

So to sum up, for many who need to examine your AI enterprise thought, to see if that’s one factor which may be solved with utilizing AI, you’ll be capable of try answering the subsequent three questions:

1. Can you particular your question in writing?

2. Can an educated human do the job?

3. Is the context on the market?

In case you possibly can reply certain to all three, then you definately’re ready to maneuver on. There ought to nonetheless be hurdles to beat, and perhaps it appears to be too troublesome in the end. Nevertheless that’s the topic of 1 different submit.

Good luck!

With sincere regards
Daniel Bakkelund