Artificial intelligence to sale

The purpose of this article is to try to present a more posed and pragmatic approach to the artificial intelligence market, seen by a person who started AI in 1991 before going through a winter of Artificial Intelligence for 20 years. 
I suggest to address several points:
  • What is artificial intelligence?
  • NOAI: Not Only Artificial Intelligence
  • Some business cases
  • What to sell
  • What prevents us from selling AI
  • About resources and staffing
Of course, I do not pretend to bear an absolute truth. I present here a point of view, which I hope as clearly as possible, and invite you to a substantive debate.
As we should do on any social network, if you appreciate my article and/or you want to help me, please like and share this article !

What is artificial intelligence?

teach to the computer to learn by itself! Share knowledge! Create a social contract between AI and humans

Basic principle

Wikipedia says "Artificial intelligence (AI, also machine intelligenceMI) is intelligent behaviour by machines, rather than the natural intelligence (NI) of humans and other animals. In computer science AI research is defined as the study of "intelligent agents": any device that perceives its environment and takes actions that maximize its chance of success at some goal.[1] Colloquially, the term "artificial intelligence" is applied when a machine mimics "cognitive" functions that humans associate with other human minds, such as "learning" and "problem solving".
What is more pragmatic is that artificial intelligence is a way for the computer to explore and learn strategies to provide more accurate results by continuing to improve over time to IT platform and by sharing this knowledge between human and computers: like the automatic car that would kill his passenger and share his failure to allow other cars not to make the same mistake.

AI is old

AI is a predictable evolution because old and mature for a long time
What's new is not AI, the basics were invented 20, 30, 50 years ago or even more. What is new is that we can industrialize the AI and use it in real life, thanks to the big data and the computing powered by GPU (Cuda), NLU etc...
AI is a major revolution because it will bring a lot more than IT could provides today
Here we are faced with a paradox: Artificial Intelligence is a mature domain (a predictive evolution) and by its industrialization an innovation (a revolution by its socio-economic consequences).

Fantasies around AI

Some think that around 50% of white collars will loose job because of AI. It is actually a misunderstood conclusion of a study published in 2013 that took into account only a certain category of employment only in the USA ... It should also have been interested to know more who was the main author. Any way....Today we speak of less than 5% of impact and we tend to zero: AI like all the major industrial revolutions, changes the deal in depth but redistributes all.
We do not know ! We can only hope but objectively we can not know
The point of singularity would be that moment when the AI becomes intellectually independent able to develop new concepts by itself. If this has been the basis for magnificent science fiction works, it has become the business of some self-proclaimed gurus, in the vast majority non-practicing of AI, or a way to over-sell AI playing on fear (FEAR = PERFORMANCE so BUY). But when we look in detail this subject, experts realize they dont know ... Ok it's less sexy for TV debates, ...and to sell conferences and books.

Specifically for IT

The AI is first and foremost a way to improve the accuracy of the IT solution (prediction, maintenance, decision making, etc.) and to be able to control systems where modeling is impossible because the reality is too complex for to be precisely modeled.
AI is also used in the field of man-machine interfaces:
  • Face and expression recognition
  • Speech Recognition
  • Natural language dialogue
  • Bot and Robot
AI is also a market at 50% growth per year ...

NOAI - Not Only Artificial Intelligence

AI alone is not enough!
A real design on my whiteboard mixing Machine Learning, decision tree, etc...
Artificial intelligence is just another possibility, and is not an end in itself. The approach that seems to me the most mature is to always associate the AI to deterministic mathematics.
For example in the field of decision-making, if you have a great history (data), machine learning is an excellent solution. But only you will have a consensus that will allow you to automate your classifications (decision), but not to improve them qualitatively. In my approach I will rather associate machine learning with techniques like multi-criteria methods to improve the scoring of the data that will be used for my machine learning: The better it is trained, the better it will be.
Let's not forget that AI does not serve everything. If we can avoid it, it's better.

Some business cases

AI is set for everything ! the question of business case seems to me secondary. The question is not whether we should use AI, but where and when!
We'll go on the bots and humanoid robots that I'll treat another time, but I give you some cases where the AI is a real plus:
  • Predictive maintenance (anticipate the most precise maintenance interventions): IT support, industrial maintenance, etc ...
  • Creation of knowledge base and advanced search engines for portals like LiveLink, SharePoint, LifeRay, etc ...
  • Decision on credit and insurance: AI usable to improve decision trees, and the weighting of criteria, creation of ecosystem score, etc ...
  • Negotiation (price sensitivity, negotiation break, choice of third party, etc ...)
  • Fraud and security: Detection of fraudulent behavior and return to a honey pot, inconsistency analysis, etc.
  • Marketing: Analysis of social networks and feelings.
  • Prediction crime, electronic intelligence
  • etc.
We can distinguish in fact two categories of applications:
  • Applications functionally enhanced by AI
  • New applications born with the AI
In the end the question of the business case is not so relevant that: as soon as a human can improve an application by changing parameters or request a "change" on an application to developers, the AI is a possibility to evaluate, and potentially a plus.

What is actually sold(able)

In fact we do not sell AI ... You all know the story of the sage who shows the moon with his finger. Idiots look at the finger. The AI is the finger.
What we have to sell are the augmented solutions made possible by artificial intelligence, not the AI itself.
The paradox is that the AI part of a project is never what generates the most line of code ... Machine learning can be delivered with less than 200 lines of code. The key is in capturing the business intelligence and data needed to implement this intelligence.
Another point is the man + AI IT alliance. How we create a kind of social contract between the human and the AI:
  • Allow the human to learn from AI
  • Enable AI to learn from humans
  • Allow expert knowledge to be easily consumable by human beginners, and help them progress faster.
In terms of project management and pre-sales this has some consequences:
  • We must start with a capture of human intelligence (individual, collective) for example with an assessment.
  • See how to create a real alliance between humans and AI as much for reasons of acceptance as business efficiency.
  • To propose a minimum qualitative result, knowing that when one sells a system that learns, it is impossible to know until where he will learn.

What prevents us from selling AI

AI is a real elitist profession: When you consume AI without really understanding AI you do not AI
There are many factors that slow the development of this market.

Get started easy

With publishers such as IBM or Google, the AI seems easy. And to a certain way it's true: Watson for the extraction of sentiment (tone analyzer) or the extraction of entities (NLU) ... But this is to consume artificial intelligence.
This accessibility is based primarily on a platform dependency, be it google or IBM. Except that by definition intelligence is the ability to create new concepts by associating concepts not nearly related. It is the same for AI, we must compose solutions by combining different elements AI and NOAI (remember : NOT ONLY ARTIFICIAL INTELLIGENCE).
Of course, the next step is the development on frameworks such as Google Tensorflow ... Here again it's easy ... Very quickly it is possible to create a first solution. But it's not because Tensorflow allows you to do Deep Learning that when you do Tensorflow you have to do Deep Learning.
We can conclude two things:
  • Many people think of doing AI while they only consume it without really understanding it.
  • This has opened up opportunities for people who do not really have the skill, and who will be disillusioned and disillusioned with customers.

The Next Step

Watch out for the step ... If the beginnings are easy everything is accelerating and faster and faster. And with the maturity of customers, the bar will be much higher, and knowing how to consume AI will not suffice. We will have a need for real skills and a real understanding of the subject, as much for sellers, analysts, as architects and designers of AI solution.
The problem is that with the winter of artificial intelligence, we have not enough people really trained: a kind of data scientist x 1000.

The nature of IT management

With the frameworks, and the very strong industrialization of IT, and as in other areas we have seen a hyper politicization of management, in the true sense of the word, with the consequence of a partial loss of operational knowledge and then a loss almost total of technical competence, which was previously non-impactful and ultimately normal, as the business did not require arbitration on complex technological issues. Except that the AI with its transversal aspects, not determined by the deliverables, and its own complexity and also contextual - AI is booming - means that the managers involved must acquire new skills, and have a technological and strategic watch to be able to handle this at best.

Paradox of the lack of competence

The paradox is that this lack of available skills associated with managers who are in the vast majority not trained enough to AI (which is the heart of the digital transformation) blocks the development of AI while technologies are mature , the available material, and customers ready ....
But if we take this into account, it also means for the teams who understand this, there are incredible opportunities ... Provided to put the means in knowledge (Intelligence becoming the most important capital of a company - I invite you to read Jean Tirole, Nobel Prize in Economics 2014).

Impact on resources

Let's list the different roles:
  • The designer / architect AI (NOAI): A kind of Data Scientist that includes artificial intelligence first. He designs solutions by combining different AI and non-AI components. He selects the tools, the framework, and describes the specifications of the developments to be made.
  • The "AI coder": he develops custom-made AI, mainly based on existing frameworks and tools, but also by typing custom code.
  • The AI Business Consultant: he collects and analyzes the individual and collective intelligence of the client to define the business needs and the nature of the human IA contract.
  • And all the usual IT professions: AI is only an extension of IT ....
Because the AI is mature and confirmed for decades, and customers are starting, the vast majority of customers have the same technical needs: the difference being made on the choice data, and intelligence to implement.
For a company like the one that employs me, just a few designers and some AI coders to provide all the necessary components for all teams are needed. It remains to train AI consultants and developers to choose and consume the right components, making industrialization the key.
When it comes to the client's impacts, a change management is a critical point !!!

Conclusion

WE DO NOT SELL ARTIFICIAL INTELLIGENCE! We sell business solutions boosted by artificial intelligence.
While nobel prices, scientists and economists are talking about a major industrial revolution, the IT world that is the bearer of this revolution is still struggling to get to the technical level necessary for many reasons, and yet some companies, some teams understood the ins and outs.
What is important to remember is that AI is primarily a complicated business when you manufacture components from A to Z, requiring knowledge in many other fields, and is complex in the association and construction of design. For a few years, this is still an elitist domain, where the most efficient project management mode is XP (eXtrem Programming).
Complex but not complicated for 99% people
When we have the right team, everything is possible, and a good AI project is a small project (less than 100 days).
If this article interested you, I invite you to "Like", to share, ...
Jerome Fortias - contact@jfortias.net
PS : This article was translated by an AI powered solution :-)
Suivez moi / Follow me sur/onTwitter https://twitter.com/FortiasJerome 

Commentaires

Posts les plus consultés de ce blog

Quand Google fait un KIT AI hyper débutant !!!

50 niveaux de RaspBerry Pi

Netflix AlphaGo - le documentaire à voir absolument