2023 can be called the year of AI, especially given the hype around ChatGPT. But is AI really a panacea? Will he be able to put people out of work? Let's look into this issue.
Friends, if you don’t like to read, then below is a video recording of our webinar and a presentation for it.
Content
What is artificial intelligence?
So what is AI?
Artificial intelligence is any mathematical method that can imitate human intelligence. The first projects begin in the 1950s.
There are 3 key areas here.
Neural networks are mathematical models created in the likeness of neural networks in the brains of living beings. The most popular approach to creating AI now.
Machine learning (ML) - statistical methods that allow computers to improve the quality of the task they perform with the accumulation of experience and additional training. That is, we are talking about how the neural networks of living organisms work. This direction has been known since the 1980s.
And deep learning (DL) is not just about teaching a machine with a human telling you what is true and what is not, but also about systems teaching themselves. This is the simultaneous use of various training and data analysis techniques. This direction has been developing since the 2010s.
Weak, strong and super strong AI
Now about 3 concepts - weak, strong and super strong AI
All that you and I know now is weak AI. Weak AI (ANI, Narrow AI) can solve highly specialized tasks for which it was originally designed. For example, it can distinguish a dog from a cat, play chess, analyze videos, improve video quality, etc. But, for example, the most powerful AI for playing chess is absolutely useless for playing checkers
Strong (AGI) is AI that can navigate changing conditions, model and predict the development of the situation. And if the situation goes beyond standard algorithms, then find a solution yourself. For example, solve the problem “go to university.” Or learn the rules of the game of checkers, and start playing checkers instead of chess.
What qualities should AI have?
Thinking is the use of methods such as deduction, induction, association, etc., which are aimed at isolating facts from information and presenting them (preserving them). This will allow you to more accurately solve problems under conditions of uncertainty.
Memory - the use of different types of memory (short-term, long-term). Can be used to solve problems based on previous experience. Even if you try to communicate with the newest version of ChatGPT 4, you will see that the algorithm has a small short-term memory, and after 10-15 messages it forgets first how it all began
Planning - tactical and strategic. Yes, there are already studies that claim that AI can plan its actions and even deceive a person to achieve its goals. But now it is still only in its infancy.
Learning is imitation of the actions of another object and learning through experimentation. Currently, AI learns from large amounts of data, but it does not model or conduct experiments itself. Although, we do not fully understand how Chat GPT works. And this is one of the main problems
And now we are at the stage of transition from weak to strong AI. Yes, ChatGPT from OpenAI, LaMDA from Google can generate text through query analysis and big data processing. But they only broadcast what their creators taught.
And truly strong AI (according to various estimates) is still 5-20 years away.
Also, various researchers identify a level of AI – super-strong AI (ASI). This is AI that can not only solve complex problems, but also do it almost instantly. If there are hundreds of weak AIs for each task, there will be dozens of strong AIs (most likely there will be a division in areas), then there will be 1 superstrong AI per state.
In general, in our opinion, the development of strong and super-strong AI may take a long time. And there are several reasons for this.
Firstly, this is a very difficult, costly and risky task. The era of uncontrolled AI development is ending. More and more restrictions will be placed on it . As a result, strong and super-strong AI in a risk-oriented approach at a high level of risk. This means there will be many more restrictions. Weak models with narrow specialization will remain more “free for development.” Secondly, this is simply impractical. Two key trends in the development of AI:
creation of cheap and simple local models for solving specialized problems;
creating AI orchestrators that will decompose a request into several local tasks and then redistribute this between different local models.
As a result, we have a simpler, safer and cheaper solution to problems than creating strong AI. We wrote more about key trends here .
Where can AI and machine learning be used now?
The most relevant areas for the use of AI with machine learning can be identified as
forecasting and decision making
analysis of complex data without clear relationships, including for forecasting
process optimization
pattern recognition, including images and voice recordings
content generation
And the key area that is now at the peak of popularity is image recognition, including images and voice recordings, and content generation. This is where most AI developers go.
Examples of such services:
video - supercreator.ai, tavus, windsor
images - stockimg.ai, midjourney, dreamer
text - chatGPT, notionAI, jasper
research - bearly, scholarcy
design - looka, galileo AI, uizard
presentations - slidesAI, murf AI
audio - whisper memos, soundful, steno, podcast.adobe
At the same time, the combination of AI + IoT deserves special attention
AI receives clean big data, in which there are no human errors to learn and find relationships
The efficiency of IoT increases as it becomes possible to create predictive analytics and early detection of deviations
How AI models are trained
Most AI models are trained in the same way as children are trained: they are given information as input, the neural network gives an answer, and it is told whether it answered correctly or not. And so time after time. This type of learning is called “supervised learning.”
At the same time, there is also unsupervised learning, in which the system spontaneously learns without human feedback.
Disadvantages of Current AI Solutions
Amount of training data
Neural networks are demanding on the quality and quantity of source data. But this problem is being solved, but nevertheless they still need a lot of labeled and structured data
Dependence on data quality
Any inaccuracies in the initial data greatly affect the final result
Ethical component
Who should the autopilot shoot down in a hopeless situation: an adult, a child or a pensioner? There are countless similar disputes. For artificial intelligence there is no ethics, good and evil. There is also no concept of “common sense” for AI.
Neural networks cannot evaluate data for reality and logic
Neural networks simply collect data and do not analyze facts and their connections
Quality of "teachers"
Neural networks are taught by people. And there are many restrictions here: who teaches what, on what data, for what?
People's readiness
We must expect enormous resistance from people whose jobs will be taken away by neural networks.
Fear of the unknown
Sooner or later, neural networks will become smarter than us. And people are afraid of this, which means they will slow down development and impose numerous restrictions
Unpredictability
Sometimes everything goes as planned, and sometimes (even if the neural network does its job well) even the creators struggle to understand how the algorithms work. Lack of predictability makes it extremely difficult to eliminate and correct errors in neural network algorithms
Restriction on type of activity
AI algorithms are good at performing targeted tasks, but are bad at generalizing their knowledge. Unlike humans, an AI trained to play chess will not be able to play another similar game, such as checkers. Additionally, even deep learning does a poor job of handling data that deviates from its training examples. To effectively use the same ChatGPT, you must initially be an expert in the industry and formulate a conscious and clear request
Costs of creation and operation
It takes a lot of money to create neural networks. According to a report from Guosheng Securities, the cost of training the GPT-3 natural language processing model is around $1.4 million. Training a larger model may require as much as $2 million.
If we take ChatGPT as an example, then just processing all requests from users requires more than 30,000 NVIDIA A100 GPUs. About $50,000 will be spent on electricity daily. A team and resources (money, equipment) are required to ensure their “life activity”. It is also necessary to take into account the costs of engineers for support
What everyone strives for
Machine learning is moving towards an ever-lower barrier to entry
Very soon it will be like a website builder, where basic use does not require special knowledge and skills. The creation of neural networks and data science is already developing according to the “service as a service” model, for example, DSaaS – Data Science as a Service.
You can start getting acquainted with machine learning with AUTO ML, its free version, or DSaaS with initial auditing, consulting and data tagging. Moreover, even data marking can be obtained for free. All this lowers the barrier to entry
Everyone is striving to create neural networks that require less and less data for training.
A few years ago, to fake your voice, you had to provide a neural network with one to two hours of recordings of your speech. Two years ago, this figure dropped to several minutes. Well, in 2023, Microsoft introduced a neural network that only takes three seconds to fake.
Plus, there are tools with which you can change your voice even online.
Creation of support and decision-making systems, including industry ones
Industry-specific neural networks will be created and the direction of recommendation networks, so-called digital advisors or solutions of the class “decision support and systems (DSS) for various business tasks” will be increasingly developed.
Case Study
There is a problem in project management - 70% of projects are either problematic or failure
the average exceedance of planned deadlines in 60% of projects, and the average exceedance - 80%
budgets were exceeded in 57% of projects, and the average exceedance was 60%
failure to achieve success criteria in 40% of projects
At the same time, project management already takes up to 50% of managers’ time, and by 2030 this figure will reach 60%. Although at the beginning of the 20th century this figure was 5%. The world is becoming more and more changing, and the number of projects is growing. Even if we look at sales, they are becoming more and more “project-based”, that is, complex and individual.
What do these project management statistics lead to?
Reputational losses
Penalties
Margin reduction
Limiting business growth
The most common and critical errors are:
unclear formulation of goals, results and project boundaries
insufficiently developed strategy and project implementation plan
inadequate organizational structure for project management
imbalance of interests of project participants
ineffective communications within the project and with external organizations
How do people solve this problem? Either they do nothing and suffer, or they go to study and use task trackers
As a result, after analyzing our experience, we came to a digital advisor - artificial intelligence and predictive recommendations “what to do, when and how” in 10 minutes for any project and organization.
As a result, project management becomes available to any manager for 1000-2000 rubles per month
The AI model includes a project management methodology and sets of recommendations. And the neural network will prepare sets of recommendations and gradually learn on its own and find new patterns, rather than being tied to the opinion of the creator and the one who will train the model in the first stages.
The neural network will prepare increasingly “complex” recommendations with each project in order to ultimately increase the level of manager competencies and the organization’s project culture.
Conclusion
AI is currently on hype, disappointment will follow in any case, but don’t be afraid of it, all technologies go through this
At the same time, AI is now only in its infancy. The real heyday of AI will be in 5 years, when it becomes less demanding on the quantity and quality of data
Recommender systems have the greatest potential; for this you need to learn how to decompose business processes into components. Such systems will provide the greatest benefit to small and medium-sized businesses
Nowadays, a huge number of specialists go specifically to image recognition and content generation, while there is a shortage of specialists capable of building recommendation systems
Artificial intelligence will not take away a person’s job. But a person who knows how to use it will do it easily
AI will only replace low-skilled labor. Some specialties will die, but new ones will appear
AI is moving towards model builders similar to website builders
AI will become the driver that will move us towards further development, just as the wheel, steam engine, and industrial conveyor did in their time.
The main thing is that artificial intelligence is one of the tools of a systematic approach that works in synergy with other tools.
***
We develop our own digital solution for your projects. You can get acquainted with it at the link:
Additional materials
Video
Links and articles