Read the fine print
Artificial intelligence marketing over-enthusiasm is pushing us down a dangerous path. We hear about “massive amounts of documents ingestion”, “barriers-braking computer processing power” without thinking of purpose nor content of these documents.
Recurring failures of Business Rules Engines projects illustrate uselessness of trying to inject expertise into tools the same way code is done. After years of project, over-complex rule sets are generated, hardly maintainable. These projects most of the time ended in “our experts will never agree, let’s stop this” . AI projects just now go the same. As if bringing more and more experts in the past would have made the Business Rules Engines projects work.
Observations and facts indingestion, is demonstrated by the well known confusion matrix in machine learning : If adding more and more information creates more experience, it is far from creating expertise and can even result in worse results with time.
Making a simple analogy, the best diet is of course not eating more junk food, but eating the right thing corresponding to your activity.
Know the unknown
It becomes clear that tooling is not the issue, nor solution. Human knowledge, expertise and role are in fact this food that fuels all business. And it follows a lifecycle, within which every single step is mandatory.
It is critical for a company to understand its own operating model and assess real difficulties and strengths.
Today, AI use cases can be split into two main categories : somehow enhancing user experience (personal assistants, chatbots…) or optimizing processes (RPA, contact center automation…).
Regarding this last category, it is usually hard to understand that ongoing initiatives are going the wrong way. They try to cure the symptoms (cost drifts, lack of efficiency), not the illness itself (lack of real expertise, conflicting drivers between business units…). Feeding diabetic with more and more sugar, company is replacing experts with expert systems, and agents with chatbots.
I try to give an overview of all these different types of Artificial Intelligence use cases in another post.
Distill appropriate level of knowledge
Understanding the appropriate diet and cropping the good food accordingly is the key. Of course tools will come into account in the end : Overwhelming recurrent jobs will be supported with doers tools fed with facts. Small experts pool set will substantiate their expertise with helper tools and, in the end, be able to efficiently share not facts but real knowledge…
It is also important to clearly identify the expected outcomes of transformation. Beyond vendor marketing, there is of course no “one fits all”. Auditability can be a key element. And statistically giving a result based on huge amount of facts if far different from being able to explain underlying patterns.
Hopefully now some families of tools will facilitate this tedious process. Facilitating is not doing the job though. And this is where most of the difficulties are.
Accept the unknown
A relevant AI use case will challenge part of what makes the core business value of the company. It can lead to delegate knowledge of market, clients but also of best practices and expertise to some kind of tool. It might end-up in being used to propagate company culture and business values to customers. All these items on which the company had the feeling to have a grip.
But were all sales agent having the right attitude with clients? Were all the contracts sold compliant with regulation? Was everyone delivering its best? ... It can even challenge the experts, putting them in front of their contradictions or pushing them to give their recipes instead of cooking with their « gut feeling ».
Appropriately used, AI is not aimed in replacing humans, but in giving back human its right place within the enterprise. This leads to strong prerequisites, from employees and from top management. First, get ready to replace processes with trust, break existing barriers. Second, proactively start working together with employees and their representatives.
Knowledge is a process of pilling up facts; wisdom lies in their simplificationMartin H. Fischer
Artificial Intelligence projects are absolutely not IT projects. Of course being able to understand all solutions philosophies and constraints is mandatory. But if solution based projects are interesting in terms of future technical operations assessment, they are close to useless in terms of real company impact.
As long as pain points of all stakeholders are taken into account, artificial intelligence can become a fantastic tool for high value use cases in the enterprise. It can become an appropriate method :
- to identify best opportunities,
- to understand the impacts,
- preferably to anticipate them
But accordingly setting the cursor between optimization, transformation and disruption is critical.
AI need to help refocusing on enterprise value, getting in capacity of doing appropriate analysis and taking good decisions. To do so, the artificial intelligence system will need to continually learn from its interactions with humans. And humans will need to learn from AI. Data needs to be consolidated, bringing digital architecture needs (interactive, coordinated and elastic). But above all, business experts role is to be redesigned. Analysis and substantiation become the key function instead of execution. Indeed, recurrent practice does not mean good practice. As does Microsoft Tay intelligent chatbot demonstrated in successfully becoming a racist misogynist within 24 hours by discussing with, and learning from internet users. Autonomous learning is not there yet.
Disrupting competitors having far more fact gathering or AI technical capabilities are there. Even Uber could be “uberized” (cf. Arcade city)!
There is no cookbook yet to go forward but all the ingredients are there. You just need to be ready to “taste and learn”.
What's up everyone, it's my first pay a quick visit at this web site, and piece of writing is in…
Wow! This blog looks exactly like my old one! It's on a completely different topic but it has pretty much…