Artificial intelligence (AI) is advancing at breakneck speed, with new breakthroughs and milestones being achieved on a regular basis. It is already capable of sifting through massive amounts of data and making objective decisions. So, what comes next?
Table of contents:
- The concept of General AI
- Narrow AI (or Weak AI) – what do you need to know?
- Narrow Artificial Intelligence – what are the most popular applications?
- Weak AI – examples
- Narrow Artificial Intelligence – wrap up
In fact, AGI has a long way to go before it becomes a reality. But first, let us define it and explain why we are pursuing it.
The concept of General AI
Artificial general intelligence (AGI) is also called strong AI or deep AI. The concept refers to a machine or system that can think and carry out decisions and tasks without the involvement of a supervisor, being a substitute for human intelligence. Its mind allows it to think, understand, and learn. It can also transfer those skills to connected machines, teaching them to recognize human needs, emotions, beliefs, and thought processes rather than replicating or simulating them.
At this point, it should be clear that making machines conscious necessitates a diverse set of cognitive abilities. Obviously, pop culture and media have already heavily influenced this plot, scaring us with images of emotionless, rebellious humanoid robots who have decided to rise up and exterminate everything and everyone [including the human race]. Alternatively, a malicious supersystem, which deceives and traps its own creator as some kind of a personal vendetta.
In reality, the concept is far less frightening and far more practical. AI systems will be able to perform tasks more efficiently than humans – and effortlessly complete higher-level tasks – but their programming will limit them to the function that has been assigned to them. As a result, they will not decide whether or not to perform a task. Or, if they wish to solve complex problems.
The goal of General AI is to make the machine/system cognitive. Similarly to the abilities of the human brain, it should have a general experiential understanding of our surroundings. Like us, it will be capable of multitasking, recalling, and memorizing information. At the same time, it will be able to process data much faster than we, which will be critical in risk-based decision-making that saves money, resources, or human lives. Plus, uninterruptible computations will most likely skyrocket research.
Today, general AI appears to be more of a sci-fi concept. Many people find it difficult to believe that a highly intelligent system will outperform humans in terms of knowledge, full cognitive abilities, and processing speed. Many researchers, however, believe that AGI is decades away and that it will not be seen this century.
Nonetheless, while the timing of AGI is unknown, the disruptive effects on society cannot be overstated. Let’s have a look at the current state of AI, represented by Weak AI.
Narrow AI (or Weak AI) – what do you need to know?
Narrow AI (ANI) refers to AI systems that are designed to perform a single or limited task. It’s the polar opposition of multitasking Strong AI as ‘weakened’ algorithms complete programmed functions only. In other words, narrow AI models can only complete tasks without the assistance of humans if properly trained. As a result, they must learn on their own through a series of (usually supervised) trainings, which takes time and money.
When it comes to the actual state of Narrow AI technologies, they are currently not able to achieve self-awareness. According to the McKinsey listing, they lack sensory perception, natural language understanding, social and emotional engagement, creativity, motor, navigation, and problem-solving skills in order to be fully developed.
Narrow AI systems “think” by completing or solving a set of pre-defined functions that have been taught to them. Their ultimate goal is to increase accuracy. ANI has no memory [hence, no reference to historical data], so the knowledge from each task must be manually reapplied to other tasks. Furthermore, it focuses on simulating rather than deducing.
The vast majority of current AI applications start with weak AI. This type of artificial intelligence can be thought of as a foundation for neural networks that simulate sentience or consciousness.
Let’s now move on to the most popular applications of ANI.
Narrow Artificial Intelligence (ANI) – what are the most popular applications?
Improving the diagnosis accuracy with computer vision
Narrow AI has the ability to focus on a single task and perform it far better than humans. It does not feel tired, and when properly trained, it can detect any microscopic change in the compared images. For example, if you feed a deep learning algorithm enough images of skin cancer, it will eventually be able to detect it better than experienced doctors [learn more about our case study].
Bots: Increasing Customer Engagement
Bots are an excellent example of narrow AI in action. A bot is a piece of software that can automate simple and repetitive tasks. It can provide consistency, accuracy, and speed in customer interactions.
Unlike human beings, bots won't get tired or frustrated while going through the same conversation multiple times a day. They will always follow a recent version of the script tree and are guaranteed to behave consistently with any type of client. They can easily look up product details, shipping dates, order histories, and so on in the knowledgebase.
They can also communicate in any language – which is also their biggest limitation, as some translation engines (e.g., Google Translate, DeepL) are still unable to capture the exact meaning of the translated sentence. Understanding language (i.e.: more detailed questions from customers) is a general problem for the system, which Natural Language Processing is attempting to solve.
The majority of jobs can be automated with narrow Artificial Intelligence, enabling humans to handle parts that necessitate human care and attention. Take for example previously mentioned customer service. While bots filter out most general issues (which can be solved by pre-defined rules: scripts), humans can concentrate on more complex cases. This component could be extremely beneficial to any type of business that works with a vast database. As a result, customers are less frustrated, wait times are shorter, and employees' time is better utilized.
Weak AI algorithms can help teachers identify and close learning gaps. Narrow AI algorithms are still a long way from developing the complex social skills that teachers require to guide students toward knowledge. They can, however, help teachers improve their craft.
Weak AI – examples
- Chatbots and conversational assistants. Virtual assistants such as Google Assistant, Siri, and Alexa are the most well-known examples of Narrow AI. The same category applies to simpler customer-service chatbots, such as those that assist customers in booking vacations.
- Software for facial and image recognition. ANI is widely used by tech giants such as Google, Apple, and Facebook. The technology assists in identifying people [or objects] in photographs by suggesting personalized clusters ['memories'] or similar outcomes (i.e. search results).
- Video/Music recommendation services. Similarly to the preceding example, Weak AI can analyze platform user behavior and recommend similar results based on selected artists, genres, tags, and general trends. The more views/playtime a result receives, the more accurate it becomes.
- Speech Recognition.
- Spam Email Filtering.
- Predictive maintenance models. ANI models can predict when a machine part will fail by analyzing data collected by sensors and informing users ahead of time.
- Autonomous vehicles. Vehicles that operate autonomously or semi-autonomously are another great example of narrow AI applications. This category also includes semi-automated drones and factory robots.
Narrow Artificial Intelligence – wrap up
As machine learning becomes more integrated into modern society, narrow AI applications are becoming more common. However, there are some reservations about using ANI in critical infrastructure functions on a large scale. Some argue that Narrow AI is brittle due to its characteristics and that alternatives may be more risk-averse in cases where a neural network is used to control critical systems (e.g., hospital systems, power grid, financial trading).
However, we must keep in mind that the same systems controlled by humans are not much more failproof – and the possibility of excluding human error and establishing another pair of eyes may be especially beneficial in medicine, data-driven decision-making, and improving cybersecurity.
AGI may not be ready this decade, or even this century, but some of its capabilities may start to emerge in unexpected places. And, when the second pair of eyes is transformed into the second brain, we will be in the midst of General AI, which, in addition to advising, will almost certainly suggest a much more efficient way to solve upcoming problems. Until then, narrow algorithms designed to help machines complete tasks will continue to change our lives.
And remember, those who are vigilant and well-prepared will benefit the most. It's best to remain one of them by staying current on tech news.