Ideas of AI pricing/business models

The AI becomes to a hot topic nowadays, and not only for scientists or for visionaries. More and more industries are started to look to an AI direction and thinking how they can use this new technology for their business.

But what kind of business models we will see in near future, let’s do some looking into a crystal ball.

Infrastructure for AI

AI (Artificial Intellect) and ANN (Artificial Neuron Network) is very resource consuming. So, therefore, it is understandable that the main need at the moment is right hardware. Existing AI approaches requires a lot of small floating point calculations. It was a little bit surprise for me, but if we look deeper it makes sense, the most demanding (from the resource point of view) are computer graphic who also use floating point calculation to create VR (Virtual Reality).

That’s why the biggest name in a game now start to change their position in the market. NVIDIA are not only the biggest player in GPU (Graphics processing unit) market, but also build hardware solutions for AI. Even now a lot of AI enthusiast use NVIDIA graphics cards to make AI calculation, and I read a topic where author purpose to change NVIDIA slogan from “The Way It’s Meant to be Played” to “the AI computing company”.

So there is a place and requirements for hardware to run AI in data centres or organization private cloud solution. I believe this model will be quite similar to existing models in Data centre and hardware market.

Neurons network — price per neuron

A capability of AI mostly lands on the size of neuron network. So, the most basic model will provide untrained neurons network, which can be trained and adjusted for the specific purpose. This business model is quite similar to IaaS (Infrastructure as a Service) or PaaS (Platform as a Service).

I see two main models:

1. package of inner connected neurons — block
2. on demand neurons network

Both models provide pricing per neuron. Package model will be invoiced in advance, On-demand model provides postpaid.

A client receives API to neuron network according to an agreement (like AWS — Amazon Web Service), and there can be added additional conditions (for example calculation limitation, type of neurons (RNN (Recurrent neurons network) of LSTM (Long Short-Term Memory networks)), etc). I can imagine, that for native mobile apps or cars there can be not only Cloud AI services, but also libraries for using AI autonomously.

Pretrained neuron networks — price per â€œskill”

This model provides AI to the customer with specific skill — for example, image recognition. With this basic pre-trained skills, a customer can apply AI into their process. The benefit of this model is that customer receives some external skill. As a part of the process this skill can be improved or adjusted according to customer needs by the environment input and feedbacks (for example: AI can recognize persons — we can provide the feedback which persons are unwanted, and instruct AI to issue alerts if such person is recognized).

As a skill trade — you pay per skill and how good this skill is. Lower level skills as a basic object recognition will be cheaper than a face or speech recognition.

Trained neuron network — price per “competency”

Pack of pre-trained skills to complete complex tasks. This AI can perform in a much complex environment. The first characteristics we can see in Facebook Chat Bots. In the beginning, you get bot with basic skills (text recognition, context awareness) and you can train this bot to perform for example product advertiser or basic support provider.

Price per â€œrole”

Pack of competencies to perform role — a good example is IBM Watson Office assistant. This kind of AI can replace some of the existing roles in the organization — and is quite easy to calculate price comparing price for AI vs existing human expenses.

And this will continue, there is a lot of activities to find a better way how to perform different tasks. For example: Smacc working on accounting solution. The biggest business names predicting, that in a time of 10 years they will hire an AI as a board member.

This is the scariest moment for a lot of people, but do we really have a reason to be scared? In the beginning of the previous century, the main shift happen. The invention of the washing machine freed a lot of women’s time. This shake labour market, because this flooded marked with fresh working hands. More and more women become educated and take a job where in the past been limited to men. In a very short period of time mankind figured out how to employ all. So the main lesson is that the people are very creative creatures and human will figure out how to employ themselves (or others). Some of these roles I described below.

Performance-based price

This one can be a very interesting marketing/PR opportunity for AI companies. The main idea is similar to the existing hiring process. If you are hiring junior, who can perform only a simple task, the starting salary will be small. And the similarity continues — the Junior earn some knowledge about Your processes, and become more capable to do more you lift up salary.

So the main idea is — pay by achieved results. This allows companies to pay less when the AI is learning, and pay more when AI can complete many complex tasks (become an expert or senior specialist).

AI-powered services

We already see such services, especially in an infrastructure environment. Most of largest IaaS providers using AI to manage their infrastructure and balance usage of it according to client’s needs. Another service, wildly used, google translation. Just five weeks ago Google announced the major breakthrough in their translation service, and significant improvements in translation using AI.

AI identified algorithms

The power of AI is recognized patterns. The weakness of AI is resource consumption. The highest possible is a scenario, wherein the beginning the AI is operating, to identify possible patterns, and later, if the environment does not change a lot, we use reverse engineering to create simple algorithms which can operate effectively enough. This allows us to understand new environment (finding patterns), and if we find the winning algorithm, we can reproduce it cheaper.

AI hardware

This sort of AI already exist — self-driving cars use it. Moving forward and shrinking the size of AI devices (or AI capability) we will see a lot of that kind of hardware. In CES 2017, the biggest keynote where from Nvidia. And for AI the Nvidia announced a cooperation with Audi and show the “brains” for next-generation self-driving Audi. The size of this device impressed me — laptop size device, who can analyse surrounding environment and take proper decision.

So it is natural to expect a continuous shrinking of AI hardware and these small AI devices will appear in a various daily used equipment.

Hardware for AI

AI by itself are like a brain without a body. As a living being, we have a lot of sensors — eye, ears, nose, hands. All of this “devices” feed our brain with information for analysis and decision-making. The situation with AI is absolutely identical — we need to provide AI with information.

The excellent example of an AI-powered machine is Boston Dynamic. To achieve that level of flexibility in a dynamic environment they use a lot of sensors, analyse incoming signals with AI and execute a required action. The existing interaction between man and machine is lame — we need a more flexible solution to provide AI with information and receive data from it — the neuron connection can be a solution, and when we find a way how to iterate in that level, we will see a gigantic jump in technology and possibilities.

Related services

As the AI will become more and more popular and widely used there will be a rise of related services. Let's look to some of them.

AI trainers

We can see such services already. If you see the benefit of using pertained neuron networks instead of ANN who works on data flow and identify patterns, you need to train ANN. You can do it for yourself (or a specialist in your organization) or you can order a service who will train your AI to act in accordance with your wishes.

This is not so simple task actually. To train someone you need three things:

1. understanding of the process
2. test data
3. methodology

So, this can be a really good business in near future, and the biggest players will be those who have a lot of data which can be used to learn.

Behavior programmers

Those who have not seen serial WesterLand — I recommend. if we are heading to human-like AI, the behaviour becomes the most significant characteristic. Like in the beginning of the Internet, all of the sites were similar. But not for long, the role of Web Designers arise. And the customers get required individuality in their net representation.

AI psychologist

Quite interesting possibility in a world full of human-like AI. And the main reason, why this role will appear is bad habits. We can look into not so distant past — first chatbots on Facebook. Based on content in a short period of time, they become racist. Of curse, the organization cannot act that way, and if AI performs an essential role, cannot switch it off. So — we need some sort of doctor for AI, who can analyze AI and “cure” it.

Afterword

We live in an exciting time, dynamic and full of possibilities. I wish everyone to enjoy it and not be afraid of it. And like a young kid keep your eyes wildly open to see all of these miracles around us.