In a product organisation aiming to construct AI capabilities into their services and products, there’s at all times the challenge of bringing the non-AI-literates onboard the AI train. While not everybody must be an AI expert, it’s needed to have as many as possible contributing with ideas and possibilities of exploiting the facility of AI to propel the corporate to the subsequent level. This is applicable particularly to domain experts and product people, who’re on top of the issues their services and products try to unravel, and knowing where the shoe pinches.
One challenge I actually have learned is prevalent, is the essential query of “Which problems can we solve with AI?”. A matter that’s surprisingly hard to reply when posed by a non-expert. So I actually have devised three heuristic questions you can use at any time when you’re looking at an issue, and you’re wondering “Can this be solved with AI?”. If you happen to can answer yes to all three of them, chances are you’ll end up in position to begin an AI project.
You’ll be able to consider an AI as an oracle that answers questions. What you’ve got to ask yourself about, is:
Are you able to express, in writing, the query you want answered?
That is, in fact, a test that applies to anything you would like to do. If you should do something, but you’ll be able to’t formulate what it’s you would like, you almost certainly don’t really know what you would like. Launching an AI project is not any exception to this rule.
Example inquiries to ask an AI may very well be
- Is there a dog on this picture?
- What’s going to the weather be tomorrow?
- What are next week’s lottery numbers?
All of those are well posed questions that may be asked. But not all of them may be answered, so we’d like one other test.
We will consider the oracle as a function mapping inquiries to answers:
The circle on the left comprises all of the questions, and the circle on the best comprises all of the answers. The oracle is the function sending inquiries to answers. The subsequent thing to ask oneself is:
Does the function exist?
This may occasionally seem odd, and it gets queerer still: it’s best to ask this query on a metaphysical level — is there any theoretical possibility for this function to exist? Allow us to have some examples:
We’ve all seen AIs answering the “dog in the image” query, so we all know that this function exists. We’ve also seen the weather forecast, so we realize it is feasible, to some extent, to predict tomorrow’s weather. But there isn’t any strategy to predict next week’s lottery numbers. And the explanation for that is that the lottery is rigged exactly with the goal of this function to not exist. It’s not possible. And that is what I mean by “on a metaphysical level”.
Why is that this necessary? Because machine learning (which is how we make AIs) is about attempting to approximate functions by learning from examples.
If we now have quite a lot of examples of how the function (i.e. oracle) should behave, we are able to attempt to learn this behaviour, and mimic it as closely as possible. But you’ll be able to only approximate a function that exists.
Admittedly, all of this can be a bit abstract, so I like to recommend replacing this heuristic with the next meta-heuristic:
Can a well-informed human do the job?
Still metaphysically, given all the data on the planet and unlimited time, can a human answer the query? Clearly, humans are pretty good at recognising dogs in pictures. And humans did develop weather forecasts, and do them too. But, we usually are not in a position to predict next week’s lottery numbers.
If you’ve got come this far, answering yes twice, you’ve got 1) a well posed query, and a couple of) you recognize that, not less than in theory, the query may be answered. But there’s yet one more box to ascertain off:
This one is a wee bit more technical. The important thing to the query is that the oracle function often needs more information than simply the query to search out the reply. The informed human being, doing the job as oracle, may have additional information to make a choice or produce a solution. That is what I discuss with because the context.
For instance, the weather forecast oracle must know the present meteorological conditions in addition to conditions from some days back to do forecasting. This information just isn’t contained within the phrase “What’s the weather going to be tomorrow?”
Alternatively, within the case of images of dogs and cats, the context is in the image, and no additional context is required.
The rationale why this is vital, is that after we train an AI, the AI is presented with questions of the kind
The AI then makes a guess before receiving the true answer, and over time it’s hoped that the AI will learn the difference between cats and dogs. But for this to occur, the difference have to be available, in order that the AI can learn to discover the difference. Within the case of images, this is easy — you simply must make certain the photographs are of sufficient quality to make the excellence possible. Within the case of weather forecasting, it becomes more complicated — you really must make an informed decision to what information is required to make a weather prediction. This can be a query best answered by domain experts, so you’ll have to achieve out to get a very good answer to this one.
But the underside line is: if there just isn’t enough information available for the informed human to reply the query, then there’s little hope for the AI to learn how you can answer the query also. You would like that context.
So to sum up, if you happen to wish to check your AI project idea, to see if that is something that may be solved with the usage of AI, you’ll be able to try answering the next three questions:
1. Are you able to express your query in writing?
2. Can an informed human do the job?
3. Is the context available?
If you happen to can answer yes to all three, you then are able to move on. There should still be hurdles to beat, and maybe it seems to be too difficult in the long run. But that’s the topic of one other post.
Good luck!
With sincere regards
Daniel Bakkelund