Decision Avatars
Dr. Roger Moser
Faculty, Board Member & Investor, Executive Coach / Decision Intelligence Thought Leader
After the introduction of the Decision Intelligence CUBE framework in the last newsletter, I would like to discuss today the logic and some consequences of the DECISION AVATAR?? concept. Yes, there is a "??" (trademark) attached to the DECISION AVATAR?? phrase because my company registered the trademark for our idea about Decision Avatars a couple of years back in different markets including EU, USA and China (for a selected number of sectors).
Let me explain how we came up with the concept, what it contains and what some implications are for businesses, economies and societies.
I developed the first idea of the Decision Avatar concept in 2017 when I was working on the extension of the Decision Intelligence concept and was doing research in the opposite direction of BIG Data applications --> SMALL Data; e.g. in the form of expert opinions. It was around that time when I met René Eber , then a student at the Universit?t St.Gallen (HSG) and still one of the most brilliant people I have ever met and who now works at 法国HEC管理学院 .
Decision Avatar: Early Days
In 2018, René developed the AI4VC concept (#ArtificialIntelligence FOR #VentureCapitalists ) where he developed (with my humble input) a solution that would allow venture capitalists to train an algorithm to help them sort out applications of start-ups that they wanted to have a closer look at. This algorithm would replace a junior associate that would normally do such a job by reading, filtering and prioritizing the applications based on the instructions of the senior venture capitalist. The AI4VC approach has at least two advantages over the 'traditional' (analog) approach.
First, there is a lot of information loss between VCs and their junior associates making the judgement of the junior associate quite inaccurate as compared to the VCs themselves - especially when they change or adapt their assessment criteria & weighting over time based on new insights. The AI4VC solution is able to quickly identify changes in the preference patterns of a VC and adapt the ranking to new but also prior applications (which the VC might have already forgotten about it).
Second, junior associates are not cheap. The AI4VC solution is pretty simple and doesn't a lot of initial (training) data to start working.
Along with my other ongoing work at SatSure , it inspired me to develop the Decision Avatar concept. Basically, the concept is, at first sight, very close to what you could call a #DigitalTwin of the #brain but when you drill down the logic of digital twins, the concept of creating a copy of the decision-making pattern of an individual wouldn't apply to this term. That's why I finally decided to apply the concept of an AVATAR based on today's general understanding as a digital representation of a person or entity that can interact with others in virtual environments. So, a Decision Avatar is simply the digital representation of a person's or entity's decision-making pattern. The trademark was then registered in the EU, China US etc. end of 2018.
DECISION AVATAR: Concept
In the early days of the Decision Avatar concept, executives and some of my academic colleagues were quite doubtful about the potential for real-world applications. However, over the last couple of years, my business partner Abhishek Raju and I have experienced a steep rise in applications around us.
For example, imagine you are the owner of a hearing aid shop in Switzerland and have a lot of experience (and success) in selling hearing aids to all kinds of customers because you can really help them decide which hearing aid fits different problems and 'human characteristics'. But what happens when you are sick or want to go on holiday?
You might have a sales assistant in the shop but they don't have the same level of experience and customer understanding. As a result, the customer experience significantly worsens when you are not around. However, when you train your own Decision Avatar in the form of an algorithm, it can help your sales assistant to recommend the right hearing aid to your customers whenever you are not around. All it requires is some upfront training, and an ongoing analysis of your recommendation / decision-making behaviour changes (e.g. based on new hearing aids coming out) - all nicely put into an #algorithm that can then be fed by your sales assistant with a customer's characteristics and hearing aid challenges resulting in a recommendation for a hearing aid as if you were there. Enjoy your holidays!
领英推荐
The hearing aid example is truly simple. Now imagine you apply the Decision Avatar logic to the ongoing decision-making of procurement managers, HR executives, loan officers etc. Wherever people are supposed to have #stable #decision -#making #patterns and to #adapt them when the #circumstances #significantly #change .
You might argue that BIG Data #artificial #intelligence applications can do the same but there are many (business) situations where #humans are much faster and better in #learning respectively in autonomously deciding to change their decision-making patterns as the context they operate in changes. Why?
Because BIG data AI-based applications (e.g. machine learning / deep learning) are great at identifying patterns in big data sets that often humans wouldn't be able to do. However, these solutions also need a LOT of ADDITIONAL data to identify new patterns when the situation has changed. Think of the massive consumer behaviour changes in supermarkets and online marketplaces due to COVID-19. It took these companies' BIG data-based algorithms quite a while to identify new patterns while human beings were much faster to understand that people had adapted their shopping behaviour. In addition, we expect that these BIG data-based solutions are #unbiased . This requires a careful selection of the required training data. In reality, this happens much less than we as users normally expect.
A Decision Avatar, however, is #biased by definition because any person is biased. While this might represent a challenge in some areas, the use of Decision Avatars in business and society follows a different logic and purpose than their BIG data-based 'siblings'. In fact, some Decision Avatars applications specifically focus on putting managers' decision-making into algorithms and making them comparable to their peers' algorithms in order to identify inappropriate biases; e.g. when selecting candidates for a clearly defined job profile.
DECISION AVATAR: Implications for business and society
Now let's scale this concept across functions and industries - what are the implications? First and foremost, it allows companies to capture the so far primarily tacit decision-making behaviour of experts in the form of an algorithm.
Second, if companies can 'capture' the expertise & experience of their managers (experts) in algorithms, this has severe implications on how (long) they can tap into a manager's expertise and experience in the form of their specific decision-making behaviour when they are on holiday or have even left the company. During their absence, companies could use these #shadow #DecisionAvatars to support their peers or new employees to check what these experienced managers might decide and learn from their decision-making patterns. I believe you have enough imagination to understand how such a concept can change the training & development approaches of companies.
Third, the concept of Decision Avatars at an macro-economic level implies that companies primarily need a few "decision-making experts" in each area they are operating in (the same logic applies to analysing / designing activities etc.), put their decision / analysis / design etc. behaviour in the form of patterns into an algorithm and let them regularly be updated. Then the companies can apply the algorithms to all standard decision / analysis / design situations they have and let the human experts (e.g. managers) personally take care of the more difficult situations. If you question this implication, you might want to reflect on what DALL.E of OpenAI actually does. It combines the (so far) human artwork accessilbe via the web into new patterns in the form of great paintings, logos, pictures etc. - it's not the same as a DESIGN AVATAR?? (we have the trademark on this as well) but actually the combines design avatars of all artists' work the algorithm has access to.
In fact, we (unfortunately) foresee a future where the best subject-matter experts in the world are able to offer their DECISION AVATAR?? / ANALYSIS AVATAR?? / DESIGN AVATAR?? in the form of algorithms on a marketplace and make them accessible 24/7 to all people in need or, more likely, who are willing to pay for their usage in their respective (business) context. While this is great for these experts, all others (especially white-collar workers) are facing a situation where their local decision-making might not be required anymore because the experts' decision-making algorithms can cover 80 to 90 per cent of the standard situations so that only a fraction of local managers/employees is still required.
For example, if a bank has a set of top loan officers across different regions, it can apply the Decision Avatar concept to turn their loan decision-making patterns into algorithms and apply either single Decision Avatars or a combination of individual Decision Avatars to the majority of loan requests while only some complicated cases might remain with a few or single loan officers in a branch. This is likely to reduce the need for loan officers working at a bank by 60 to 80%.
The same logic applies to many 'jobs' in what Roger Martin calls decision factories. You might want to draw your own conclusions about what it means for your current job or future career.
This brings me to the truly critical part in every decision: recognizing the inherent uncertainty and exposing to the decision-maker what might happen, instead of recommending taking an action hoping for the good. For me understanding uncertainty and exposing its possible impact to the decision-maker is the real elephant in the room.
Hello Roger, it seems that by capturing the hidden algorithms of experienced decision makers we can let junior decision makers make "currently acceptable decisions." The remaining biases, also real mistakes, are those that the more experienced decision-makers fall into. By the way, this was the original objective of AI in the early 90s for the emergency room in hospitals: letting the doctors, who are not experts in all areas, record the symptoms and get a diagnosis and what actions have to be made immediately. The underlining limitation of current AI is being unable to use logical cause-and-effect. What I like to see in an algorithm is that it is based on cause-and-effect. I know there are now research and experiments of being able to include cause-and-effect within machine learning. Once cause-and-effect is part of AI then decision-making support could identify flaws in the current captured algorithm and come up with an improved one.