AI and psychology: can algorithms and human mental processes coexist?
AI has limitations in learning huge amounts of data: human psychology can help it!
AI and psychology: can algorithms and human mental processes coexist?
AI has limitations in learning huge amounts of data: human psychology can help it!
What AI and psychology are
Artificial intelligence (AI) and psychology are two fields that can be considered poles apart. They have, however, one thing in common: human beings and especially the study of their mental processes.
Let us analyze them more in detail, starting precisely with AI.
Artificial intelligence includes technologies that mimic the functionality of human thinking. Its strengths are computational power and speed. These, in fact, enable computers to perform complex operations much faster than humans.
Thanks to all these features, AI is implemented in so many areas, from medicine to finance. It is an ever-evolving technology with the goal of developing systems that can solve more and more problems.
It is clear, then, that a future in which Artificial Intelligence is widely used could bring many benefits to people. Nevertheless, if not regulated and managed properly, AI could also cause negative consequences. One example is the use of facial recognition technology in the world.
Let us turn, now, to introduce psychology and especially how the human brain works.
Psychology is a science that studies human beings’ mental processes, behavior, and human relationships. The purpose is to understand why an individual behaves a certain way in a given situation with the goal of improving the quality of life.
Everyone behaves differently because when his or her brain acquires new information, it is processed differently and unexpectedly. Indeed, when a person learns something, he or she modifies that experience as he or she goes through it based on past experiences as well.
How learning works: human brain vs AI
After clarifying the differences between AI and human brain, it is important to delve into how information learning and consequently memory occurs.
For this, we must rely on cognitive neuroscience. This is a scientific field that combines neuroscience and cognitive psychology to study the relationship of the brain to our cognitive abilities and behaviors.
How, then, do we learn and create our memory?
We must first state that memory is a process that includes encoding, storing and retrieving learned information.
In addition to learning and remembering, our brain also tends to forget. This action, however, is not to be regarded as negative. It is, in fact, essential, because our brain needs to discard unnecessary information to make room for new information.
Neural connections change according to our use or disuse of memory.
When we retain information but we don’t use it, its neural connections weaken and gradually disappear. Similarly, when we learn something new, we make new connections. Therefore, what we learn and can associate with existing memories will be easier to remember from our brain.
Compared to the human brain, AI, and in particular machine learning (ML), can recognize patterns and trends in huge amounts of data. These would be impossible to manually analyze and learn from the human brain without forgetting previously acquired notions.
Effectively, ML algorithms collect data and use it to improve at a task. For example, to find a cow in an image such algorithm is trained with images of a cow. In this way, it processes the features of a cow by itself over time.
As mentioned before, the algorithm never stops learning, as each new input allows it to improve its accuracy in detecting cows in images. In addition, ML algorithms use various methods and shortcuts to detect the image of a cow.
They analyze all the information learned during training and find their perfect combination to solve a problem through trial and error.
Therefore, machine learning algorithms prove useful in the case of large datasets.
The more data an ML algorithm assimilates, the more effective it will be in solving the problem. Does this mean that AI can acquire huge amounts of data information without any problems?
Psychology for Artificial Intelligence: the perfect match
Our brain is able to produce explanations in any situation, even in new and therefore unstable ones. For example, to understand a new situation, the brain produces one explanation. If this is changed after learning new information, our brain generates a second and new explanation.
AI, and particularly machine learning, on the other hand, behaves differently. It relies on data used for its training; this means that it is not only based on the quantity of information but also on their quality. With good inputs the algorithm is more inclined to overcome a new situation, instead, with wrong information its output could lead to wrong conclusions.
A striking case occurred in 2008, when Google made the AI-based Google Flu Trends (GFT) web service online. It aimed to predict flu-related medical visits using big data.
Google’s algorithm mined five years of web logs, containing billions of searches. It later created a predictive model using 45 search terms as indicators of influence.
GFT’s results were far from satisfactory.
In 2009, the algorithm failed to predict the swine flu pandemic. In 2012, Flu Trends overestimated the Christmas season flu peak by 50%. Subsequently, it overestimated the prevalence of influenza by 100 over a 108-week period. [1]
According to much research, having and analyzing more data does not mean getting better analyses, just as happened in this case.
How would the human brain behave in this case?
As mentioned earlier, the brain forgets by focusing on relevant data to make predictions or actions.
Neurons communicate with each other through connections called synapses. Memories are created when these synapses are stronger.
The brain continuously organizes itself based on experiences. There is a higher probability that a memory will be consolidated if it is related to knowledge already acquired or memories already present in long-term memory.
So what if AI also adopted this approach typical of the human brain?
This is precisely what happened to address the protein-folding problem, from which Alphafold was born. It is an AI system developed in 2016 with the goal of determining the structure of a protein from its amino acid sequence.
The results that Alphafold has achieved over the years are brilliant. Much of the credit comes primarily from the nature of the data.
The system has been taught the sequences and structures of about 100,000 known proteins, which represents a tiny but relevant fraction of the proteins known to science.
From this data, more than 200 million proteins have been known up to 2022 and this number is expected to grow year after year.
The system can now predict the shape of a protein in a few minutes and with very high accuracy. [2]
The two AI systems just described show two completely different results and what is the direction to take. As we have seen with Google Flu Trends, algorithms that rely on big rumourous data do not provide accurate predictions. Instead, using a less vast but relevant dataset may be the right key to making intelligent predictions.
Psychology could inspire the future of Artificial Intelligence. Over time, will researchers make algorithms operating more and more like the human brain?
References:
-
-
- Harvard Business Review. Online version: https://hbr.org/2014/03/google-flu-trends-failure-shows-good-data-big-data
- Alphafold. Online version: https://www.deepmind.com/research/highlighted-research/alphafold
-
© Copyright 2012 – 2023 | All Rights Reserved
Author: Niccolò Cacciotti, Head of AI Department
What AI and psychology are
Artificial intelligence (AI) and psychology are two fields that can be considered poles apart. They have, however, one thing in common: human beings and especially the study of their mental processes.
Let us analyze them more in detail, starting precisely with AI.
Artificial intelligence includes technologies that mimic the functionality of human thinking. Its strengths are computational power and speed. These, in fact, enable computers to perform complex operations much faster than humans.
Thanks to all these features, AI is implemented in so many areas, from medicine to finance. It is an ever-evolving technology with the goal of developing systems that can solve more and more problems.
It is clear, then, that a future in which Artificial Intelligence is widely used could bring many benefits to people. Nevertheless, if not regulated and managed properly, AI could also cause negative consequences. One example is the use of facial recognition technology in the world.
Let us turn, now, to introduce psychology and especially how the human brain works.
Psychology is a science that studies human beings’ mental processes, behavior, and human relationships. The purpose is to understand why an individual behaves a certain way in a given situation with the goal of improving the quality of life.
Everyone behaves differently because when his or her brain acquires new information, it is processed differently and unexpectedly. Indeed, when a person learns something, he or she modifies that experience as he or she goes through it based on past experiences as well.
How learning works: human brain vs AI
After clarifying the differences between AI and human brain, it is important to delve into how information learning and consequently memory occurs.
For this, we must rely on cognitive neuroscience. This is a scientific field that combines neuroscience and cognitive psychology to study the relationship of the brain to our cognitive abilities and behaviors.
How, then, do we learn and create our memory?
We must first state that memory is a process that includes encoding, storing and retrieving learned information.
In addition to learning and remembering, our brain also tends to forget. This action, however, is not to be regarded as negative. It is, in fact, essential, because our brain needs to discard unnecessary information to make room for new information.
Neural connections change according to our use or disuse of memory.
When we retain information but we don’t use it, its neural connections weaken and gradually disappear. Similarly, when we learn something new, we make new connections. Therefore, what we learn and can associate with existing memories will be easier to remember from our brain.
Compared to the human brain, AI, and in particular machine learning (ML), can recognize patterns and trends in huge amounts of data. These would be impossible to manually analyze and learn from the human brain without forgetting previously acquired notions.
Effectively, ML algorithms collect data and use it to improve at a task. For example, to find a cow in an image such algorithm is trained with images of a cow. In this way, it processes the features of a cow by itself over time.
As mentioned before, the algorithm never stops learning, as each new input allows it to improve its accuracy in detecting cows in images. In addition, ML algorithms use various methods and shortcuts to detect the image of a cow.
They analyze all the information learned during training and find their perfect combination to solve a problem through trial and error.
Therefore, machine learning algorithms prove useful in the case of large datasets.
The more data an ML algorithm assimilates, the more effective it will be in solving the problem. Does this mean that AI can acquire huge amounts of data information without any problems?
Psychology for Artificial Intelligence: the perfect match
Our brain is able to produce explanations in any situation, even in new and therefore unstable ones. For example, to understand a new situation, the brain produces one explanation. If this is changed after learning new information, our brain generates a second and new explanation.
AI, and particularly machine learning, on the other hand, behaves differently. It relies on data used for its training; this means that it is not only based on the quantity of information but also on their quality. With good inputs the algorithm is more inclined to overcome a new situation, instead, with wrong information its output could lead to wrong conclusions.
A striking case occurred in 2008, when Google made the AI-based Google Flu Trends (GFT) web service online. It aimed to predict flu-related medical visits using big data.
Google’s algorithm mined five years of web logs, containing billions of searches. It later created a predictive model using 45 search terms as indicators of influence.
GFT’s results were far from satisfactory.
In 2009, the algorithm failed to predict the swine flu pandemic. In 2012, Flu Trends overestimated the Christmas season flu peak by 50%. Subsequently, it overestimated the prevalence of influenza by 100 over a 108-week period. [1]
According to much research, having and analyzing more data does not mean getting better analyses, just as happened in this case.
How would the human brain behave in this case?
As mentioned earlier, the brain forgets by focusing on relevant data to make predictions or actions.
Neurons communicate with each other through connections called synapses. Memories are created when these synapses are stronger.
The brain continuously organizes itself based on experiences. There is a higher probability that a memory will be consolidated if it is related to knowledge already acquired or memories already present in long-term memory.
So what if AI also adopted this approach typical of the human brain?
This is precisely what happened to address the protein-folding problem, from which Alphafold was born. It is an AI system developed in 2016 with the goal of determining the structure of a protein from its amino acid sequence.
The results that Alphafold has achieved over the years are brilliant. Much of the credit comes primarily from the nature of the data.
The system has been taught the sequences and structures of about 100,000 known proteins, which represents a tiny but relevant fraction of the proteins known to science.
From this data, more than 200 million proteins have been known up to 2022 and this number is expected to grow year after year.
The system can now predict the shape of a protein in a few minutes and with very high accuracy. [2]
The two AI systems just described show two completely different results and what is the direction to take. As we have seen with Google Flu Trends, algorithms that rely on big rumourous data do not provide accurate predictions. Instead, using a less vast but relevant dataset may be the right key to making intelligent predictions.
Psychology could inspire the future of Artificial Intelligence. Over time, will researchers make algorithms operating more and more like the human brain?
References:
-
-
- Harvard Business Review. Online version: https://hbr.org/2014/03/google-flu-trends-failure-shows-good-data-big-data
- Alphafold. Online version: https://www.deepmind.com/research/highlighted-research/alphafold
-
© Copyright 2012 – 2023 | All Rights Reserved
Author: Niccolò Cacciotti, Head of AI Department