- May 16, 2019
- Posted by: Web Team
- Category: Uncategorized
There is naturally a dichotomy between AI and Creativity because in most cases AI is not employed to aid creativity nor is it currently (in the short to mid term) able to be creative. However when we are trying to figure out how we are going to use AI, then we can bring out all the creative tools, but these tools work better having access to a creative medium.
Creative machines, not
The most commonly employed AI process at the moment is Deep Learning. Deep Learning is sometimes characterised as a bottom up methodology, it often takes a large amount of data to be effective and it is much more statistical in nature, its ability to determine causality, is at times poor. You can think of Deep Learning (like many other machine learning methodologies) as a way of finding a mapping from a set of inputs (eg pictures of cats) to some outputs – the answer or prediction (“this is 90% likely to be a cat”) or preferred value. Deep Learning in its many variants is not creative.
There is a growing amount of research which might readdress the imbalance of the strong focus on data heavy AI methodologies. This research is on top down causative or rule based methodologies, in a way returning to what AI was in the “old” days – focusing more on meaning. These top down methodologies seem like a clearer path to create cognitive systems that find causality rather than correlations, but are mostly not focusing on the machine being creative, that’s a much harder problem for machines … and for humans!.
Business as usual
In my view the most potent form of creativity to use in AI is the same sort of creative process that you might use with any ideation innovation process, not so much in the actual prep of data or construction of the algorithms themselves (although does often require intense problem solving), but in deciding how and where to actually employ AI in an organisation.
The questions that you want to be creative with might be:
How can can we find value with AI?
What creative process can we use to find value in AI?
The creative medium
Ken Robinson (Robinson, K, 2017) says (at least) two very insightful things about creativity, that you don’t often hear or see elsewhere and they are:
“Because being creative involves doing something, it will always involve using some form of media. These may be physical media, such as steel, wood, clay, fabric or food; they may be sensory media, like sound, light, the voice or the body; they may be cognitive media, including words, numbers, or notation.” (Highlight is mine)
Ken also says: “Technical control is necessary for creative work…”. He also says lots of other things about creativity which are generally well known, but don’t aren’t part of my point.
By technical control, he means that one must be able to use the appropriate media well enough if order to be creative. He is suggesting that if you want to be creative at painting, you need to be able to paint well. If you want to be creative with your voice, training your voice will give you some foundation from which to be creative. If you want to be creative in mathematics, then it certainly helps if you know mathematical notation and are aware of common mathematical proofs.
Another example of the medium is jargon. Experts will have a jargon, and it is through communicating, thinking and thought experiments with that jargon that creative projects can arise.
Similarly if we want to be creative with planning potential AI projects, that is being able to strategise, create and align it within our organisations, it helps to have some competence in the “medium of AI”, specifically those thoughts, concepts and tools that help us decide how AI can be used and how it will affect our organisation. As a manager or decision maker in an organisation, If you are deciding whether to engage with AI, where and how to engage with AI, knowing how to program python, or knowing the formula for a Gram matrix, or even what backpropogation is, in my opinion not the right cognitive tools to start with. The technical understanding of AI and machine methodologies are not required for you to dream up AI possibilities.
The foundational tools and concepts that one needs to come up with new AI projects are a matter of opinion. It is the opinion of many (as suggested above) that it is necessary to understand the mathematics and technical concepts behind AI and specific machine learning concepts in order to build AI strategies, and choose AI projects.
My assertion is that for non technical executives, managers, decision makers this focus is usually:
Too low level to be useful,
Too difficult, (and thus),
Takes up too much cognitive space
Unless we are startups we usually don’t go the other way either. A good AI engineer is not expected to:
Understand strategy and differentiation
Decide on resource allocation, cash flow and financial scenarios
Build staff capability
Worry about ethics and societal benefit
Think about community engagement
Marketing, HR and the rest
It is not that with the right teaching any person can’t learn the nitty gritty of the other domain – it is just that it makes no sense. The magic of specialisation is often what gives us commercial advantage, and teaching every employee to be an expert in everything is a recipe for mediocrity (also you wont be able to, and people don’t want to learn what is outside their “identity”).
Foundational Tools and conceptual requirements (the creative medium of AI)
As I said above the foundational tools and concepts that one needs to come up with new AI projects are a matter of opinion.
I’ve chosen a set of concepts that I think are:
High enough level
to be useful to decision makers.
When it comes to AI I think we need to know i) what are the possible challenges we want to undertake, ii) can we achieve those given a certain resource priority, iii) if the AI process needs data, do we have data or can we acquire that data and iv) how will the AI process learn. In order to engage with the third and fourth point, we need to know more about AI (the creative medium) in order to be able to come up with alternative options. Even once you have the options and you decide upon one or more the process will still be exploratory like many innovative processes.
At this point you might be champing at the bit and asking “<expletive> but whats the creative medium for AI!”. Well, prepare yourself for more expletives. If AI was easy we wouldn’t have the problem of low levels of AI adoption (at least in this country). However at a high level I think this list has some useful things that would help you be creative with AI are:
A list of examples and case studies where AI has been used in the past (I use an app for that).
An understanding of the prerequisites of the different types of AI (like data requirements).
An understanding of how each type of AI learns.
Some helpful ideation and “watch out for this” checklists.
Some idea as to how different AI and non AI components might come together in a pipeline.
An understanding of your own organisations expertise/domain knowledge and how it can be improved.
An open mind
Being creative is often about coming up with lots of ideas and testing them out, so some sort of AI ideation process could be useful for AI creativity.
An AI ideation process
Given some understanding of AI concepts, the ideation process to determine AI project options is like any other ideation process. For example you could:
Ideate what do you want, or need to do, come up with alternatives
Ideate how you might undertake some of those alternatives
Do some diligence to check if they might work
Explore those options (presumably with at least one engineer).
At a high level – you might categorise how your business creates value and how AI might add value either because you have some experience on how AI can add value for certain tasks or by using what other industries as a catalyst.
You could partition your ideas based on whether you want to automate a task or augment an employee.
The combination of AI task, the type of AI tech (eg deep learning) and focus (automate or augment) will determine aspects of AI or Machine learning lifecycle, likely resource spend, employee impact, and hopeful learnings.
The set of creativity tools (adapted to AI in a way that I think is relevant) I use are based on Ed De Bono’s creativity frameworks.
Think of the shaping factors.
Random word to create alternatives.
Exaggerate, reverse etc the shaping factors.
Do a concept fan.
Once we have some creative options to pursue we can use sequential (as opposed to De Bono’s lateral) thought and planning to finish this process.
Generally when working through a potential AI process there is an interplay between the business task and the AI to solve it. For example a new Deep Learning image recognition project (eg recognising good wines) needs a certain lifecycle that is different to “solved” tasks like recognising cats where you can just plug in a black box that will do the deep learning process. Other types of AI processes might require setting up environmental constraints in a simulation or understanding expertise and human interaction for an assistant ‘bot.
Being creative with engaging with AI in my view requires a certain foundational AI knowledge – which we could call the AI creative medium. Using this AI creative medium we can then employ standard creative tools (most of which I think all are a copy of Ed De Bono’s tools and concepts) to create a bunch of ideas or options. Each idea may have its own sub idea options, and then we can choose a bunch of ideas and specific sub ideas to pursue in a pilot project.
Robinson, K, (2017), Out of Our Minds: The Power of Being Creative