Measure the QI of aartificial intelligence it is a complex challenge, since at the root there is a problem of reliability and usefulness of the results.
This because?
First of all, there is no single definition of intelligence linked to human beings (we have already talked about it in this article). In fact, there are many fields of application and spheres of life in which people can use their abilities to adapt, understand, create and act with knowledge.
There are those who, for example, may not have a marked analytical intelligence but have an innate physical intelligence (proprioception, awareness of biomechanics…). On the contrary there are those who can be physically denied, blocked emotionally and socially, but have highly developed analytical intelligence.
Like Asperger's for example, among which Steve Jobs, Elon Musk, Mozart, Michelangelo, Bill Gates ... and even Newton appear.
This variability of different types of intelligence therefore creates a variety of tests.
Content index
How IQ tests work and why they're different
Basic tests for the QI they mainly assess logical reasoning, the ability to identify relationships between different elements and cognitive skills such as verbal comprehension, visual-spatial reasoning, short- and long-term memory, working memory and processing speed.
There are different types that go to look at these skills, some are more aimed at certain age groups than others, but a substantial difference is found in cultural factors. In fact, certain tests stand out and find a relative value precisely in this: the annulment of cultural factors.
And this is a first step in trying to develop tests suitable for artificial intelligence and therefore reliable.
If we consider that human intelligence is influenced not only by culture, education and personal experiences, it therefore becomes difficult to develop tests of QI for artificial intelligence that are independent of these factors. Precisely because most tests consider these factors.
Human IQ tests applied to Artificial Intelligence
None of the tests of the QI existing can be used 100% reliably to measure the QI of aartificial intelligence. This is because these tests were designed to be used with humans, not machines.
However, some researchers have already attempted to adapt the tests of QI to make them more suitable for machines. For example, the test of Raven's Progressive Matrices was used to test the QI of some systems artificial intelligence.
- Chat GPT: This chatbot was trained on a huge dataset of text and code. He was tested Raven's Progressive Matrices, yielding a score of 129, which is higher than the human average.
- Kosmos-1: This system of artificial intelligence was developed by Microsoft. He underwent a series of general intelligence tests, scoring high on all tests.
- DeepMind AlphaGo: This system of artificial intelligence was developed by DeepMind, a Google company. He was able to beat the world go champion, Lee Sedol.
It is important to note that these results are not a definitive measure of the intelligence of artificial intelligences. They are simply a measure of how the AI performed on specific tests.
Artificial Intelligence IQ Tests: have they been done? Exist?
The measurement of QI ofartificial intelligence it is still a relatively young discipline, and there is no scientific consensus on how to do it accurately and reliably. Despite these challenges we speak of an important research area, because it will allow us to better understand the cognitive capabilities of systems artificial intelligence and to evaluate the potential benefits and risks.
Work is currently underway to develop ad hoc methods to evaluate the cognitive abilities of artificial intelligence systems stripped of obvious symbols for the culture and systems that a human being recognizes, stripped of aspects that start from the assumption that performing these tests is a person with the ability to abstract and understand the context.
Advances in this field do not exclude the fact that measuring the IQ of an artificial intelligence is a complex problem, as we have seen. To understand the problem, try to talk to ChatGpt or with google bard. You will understand that if you try to test the cognitive abilities of aartificial intelligence, without giving precise and unequivocal coordinates, it is very difficult to have reliable results, and therefore reproducible according to an objective measure.
The measurement of QI ofartificial intelligence It also raises some ethical questions. For example, it is important to avoid using tests that are discriminatory or that favor some AI systems over others. Also consider that the different AIs themselves are developed according to precise parameters and purposes. So already defining "artificial intelligence" unambiguously is inaccurate and misleading.
From all this comes complexity.
Conclusion
So, to answer the main question of the article (i.e. whether it makes sense to measure the QI of aartificial intelligence), based on what has been said so far we can say that it makes sense. But it has it to the extent that the test is able to adapt to parameters that can be uniformed to certain types of artificial intelligences examined. Given these assumptions, the meaning it would have is to better understand the limits and possibilities, risks and benefits, of the cognitive abilities of AI.
