An artificial intelligence programme designed by a team of computer programmers at the Massachusetts Institute of Technology has scored as highly as a 4-year-old on a verbal reasoning test.
The results are encouraging for the team, who believe that this marks an important step forward in the quest to create AIs which have the power of common sense and independent reasoning, rather than being limited to a narrow, limited skill-set.
Whilst the majority of AIs are specifically designed to perform one or two specific tasks, such as Google’s search engine, which is able to to rank web pages and answer keyword-heavy questions, their capabilities are somewhat limited, unable to function beyond its assigned tasks. In contrast, the AI programme developed by researchers at MIT, when tested using the Wechler Preschool and Primary Scale of Intelligence (a child IQ test), showed intelligence and common sense far beyond the range of the average AI.
The AI programme in question is known as ConceptNet, designed by Catherine Havasi and her team at the MIT Media Lab. What marks this programme out from the innumerable other AIs in existence is its extensive use of crowdsourcing, by which it is programmed with millions of of statements describing simple relationships between everyday objects, such as ‘ice cream is capable of melting’.
The idea of ConceptNet is, according to Havasi, to provide the system with “the kind of information that everybody knows about the world but that nobody ever writes down because we learnt it too early.”
Thus, when provided with an IQ test which looks to determine its verbal reasoning ability, ConceptNet will use its extensive database of information known to be true by everyone to ‘work out’ the answer. The test covers three main areas: information, vocabulary and word reasoning. Thus, to answer the question, for example, of “what do you wear on your head?”, ConceptNet would search its database for items which correspond most closely to the keywords ‘wear’ and ‘head’.
The team were, undoubtedly, pleasantly surprised by the results of their ‘4-year-old AI’. Robert Sloan, who was not involved in the process of developing ConceptNet but who led the IQ testing process and recently presented the findings at the Association for the Advancement of Artificial Intelligence conference in Washington, said that “[he] didn’t expect to see 4-year-old performance [in the AI’s aggregate verbal IQ]”.
Havasi, whilst pleased with the results, is keen to emphasise that the programme was tested only for its verbal ability, and not for other areas such as its spacial and symbolic reasoning. However, the version of ConceptNet which was tested is an early version, programmed with just one million statements. Havasi is keen to discover how their new version – which has 17 million statements and better algorithms – would score in the same test.
Sources include New Scientist
TJC offers an extensive global network of professional & experienced multilingual translators, proof-readers and interpreters. We also have academic researchers, specialists and speakers, who are all native speakers of over 180 languages.
Equipped with the broad range of skills which our network provides, TJC Global has been able to deliver a focused and dedicated service to our diverse range of clients for more than 20 years. This is why we have obtained the trust of our clientele and enjoy a reputation of being a global leader in our field.