Hello! Can anyone briefly tell me what "artificial intelligence" is?
Nowadays, when the term "artificial intelligence" is used, it is not always clear exactly what meaning the speaker puts into the term. The widespread use of this combination of words in various contexts has blurred its original meaning. Initially, AI was defined as a system with the ability to simulate any property of intelligence, use natural languages, form abstractions and concepts, solve problems that only humans can solve, and improve itself. And it happens that the term AI is now used for systems with rudimentary feedback. All this causes confusion.
Let's look at the types of AI arranged in ascending order of intellectual ability.
All existing AI systems belong to the Narrow AI group (narrow or weak AI). Such systems are based on one of the AI methods: most often they are deep learning neural networks, but there can also be expert systems, genetic algorithms, fuzzy systems, etc. They solve one particular problem under strictly defined boundary conditions. And already in many cases, better and faster than a human.
The next step, which many artificial intelligence researchers are now moving toward, is Wide AI. The main goal of such systems is to overcome the narrowness of applicability. This goal is getting closer and closer lately. Multimodality and technologies similar to large language models will make it possible to create solutions for this class of AI.
The next stage of AI development is General Artificial Intelligence (AGI). AGI is the ability of a system to solve any task in complex environments with limited resources. To put it plainly, everything a human can do, a general AI can do as well.
And finally, strong AI. This term is often used to mean AGI. But it would be more correct to describe such intelligence as many times superior to human intelligence. The most widespread idea is that if AI reaches the human level, it will not stop there, but will cyclically develop itself, increasing its capabilities at each iteration. This is where the existential problem of humanity and strong AI arises.
Having considered the types of AI, let's move on to consider the most important technological trends of the near future and their impact on the economy and retail in particular, as one of the industries where AI technologies are among the first to be applied.
For a long time, AI systems used a single channel to operate, such as text or images. Multimodality means using raw data in different combinations. For example, you can ask questions about an image. And the AI can describe what is happening in the image in terms of the context of the text question. Multimodality is both an important step toward creating Wide AI, as well as enhancing the user experience and experience. By taking a picture in the app, you can immediately ask questions about how to use the product.
Low-code AI — a set of tools that allows applying AI and machine learning algorithms without writing technical code, just using ready-made solutions, as in a constructor. On the one hand, it reduces the requirements for professional knowledge of programmers, on the other hand, it increases the requirements for knowledge and experience in creating and applying machine learning methods. This leads to a better solution of prediction, clustering, and decision-making tasks.
These systems require no development skills at all. They allow you to create new AI models and applications as quickly as possible. They already have all the basic building blocks built in, and all you need to do is configure them. This greatly reduces the threshold of entry into AI tasks, allowing specialists in their field to solve problems without being distracted by technical details. It also allows companies to adapt faster to changes in the marketplace.
It is an approach to quickly identify, verify and automate as many business processes as possible. AI is an important tool in this approach. It requires a lot of good data, integrations, and established processes for a company to use this approach. With hyperautomation, the role of employees changes from performing tasks to managing and monitoring automated systems. In the front office, where customers are increasingly using automated channels to solve their problems, the workload of support staff is reduced, and front-line employees are able to focus on complex inquiries where human contact is very important.
Right now, there are no precise legal rules for using data to train generative AI models. There is now a growing divide among experts on this issue. Some claim unconditional copyright infringement, others that there is no copyright infringement. Lawsuits have been opened. But the final resolution of this issue may take several years. The improvement of legal norms does not keep pace with the development of technology.
Here laws have already appeared and are being improved. In Russia, for example, bylaws have recently come into force to enforce the 152-FZ reform. One should be wary of leaks of such data in machine learning. In order to avoid this, synthetic data is increasingly being used. Synthetic data is artificially generated using algorithms or computer simulations that statistically or mathematically reflect real world data. The most advanced approach is to mix anonymized and synthetic data.
Here it all depends on the problem to be solved. In any case, training is based on raw data, which is often initially distorted. And if the created system makes any decisions based on this data, then a positive feedback is included, which reinforces incorrect false judgments. One example is discrimination against women in hiring by a system trained on applicants' resumes.
Generative language models can answer any question. Many people don't like this, because people like to ask unpleasant questions, and the answers become an infomercial. Large language models are censored for this purpose. A large staff of experts monitors topics of conversation shutting down the possibility of an answer system. By changing weights, some opinions are prioritized over others. The models become biased and reflect the views of the creators, but make it difficult to use it as a universal tool.
One particular use case that is considered key to hyperautomation is chatbots. They have long been used by businesses to serve customers on Web sites, but hyperautomation will allow these applications to evolve further and become more intuitive to serve customer needs more effectively. A new breath has been given to bots by the development of large language models, most famously GPT.
It's a handy tool for many tasks, but there are pitfalls. For business, it is better to use a locally installed model with additional filters and interfaces to it.