Post by by Gianluca Maruzzella, CEO and co-founder of Indigo.ai –
The road to sentient artificial intelligence is still long. We may never get there, yet since man began studying it there has been an exhausting wait for an uncertain goal like no other before. In a way, it almost seems like you can’t wait to announce that artificial intelligence is able to act like a human being and, perhaps, destroy us all. Like the super computer Hal from a Space Odyssey, capable of rebelling against its crew and exterminating them before being deactivated.
In the last few days, we have read dozens of comments and analyzes onGoogle engineer according to which the artificial intelligence he was testing (which is called TheMDA) would become sentient. An inexplicable reaction for those who are experts in the sector: Blake Lemoine simply talked with a last generation linguistic model trained in the best way. As if Lemoine wanted to believe that LaMDA was sentient for the simple fact of being able to follow a complex conversation on different levels or because the question “are you a sentient being” the AI is able to answer “yes” and also to argue about their fears or desires. Those who work in the industry know that this AI has been trained to respond “as if it were self-aware” and evidently knows how to do it well. Lemoine probably wanted to see something in this that is not there.
What LaMDA does is certainly exceptional for the depth it has achieved, but in the substance of the facts it does nothing but listen to a question and answer as they do. Siri, Alexa and Google Home. Simple applications, but extremely complex from a technological point of view. Precisely because natural language is probably the most complicated feature for a machine to reproduce.
But given the system quality LaMDA, we cannot be surprised that his answers were pertinent: what struck the engineer, however, was the model’s ability to contextualize the answers by talking about himself, exposing himself about his rights and his own personality, as if he were aware of being a machine . A detail that convinced Lemoine of LaMDA’s ability to understand his nature and to have feelings that we attribute only to human beings.
The Turing test
However, knowing how to think and speak correctly cannot be a sufficient condition for being sentient. Also because something similar has already happened with Gpt-3, the latest OpenAi algorithm that after storing 175 billion parameters is capable of writing texts that – according to the Turing test – readers cannot distinguish from those processed by a being human. This confirms that it is not artificial intelligence that is sentient, but man who educates and programs it so that he is capable of making it believe.
If such an evolved model is trained on billions of dialogues with the aim of learning what a conversation is and how to respond in a relevant and contextualized way, it simply reproduces a path already seen – as if it were a top-level chess game. – in a way that is far from what we could attribute to a human being and whose parallelism becomes a real forcing.
Lemoine probably wanted to see something in his creature that is not there. On the other hand, the ability to feel it implies the presence of an apparatus capable of collecting stimuli from the outside. But even if we consider electrical sensors as an apparatus suitable for the collection of stimuli, we should recognize that LaMDA does not have any. This is why calling it sentient is impossible. Rather, we should recognize that this was the case well trained to emulate emotional conversations so real that the engineer called to test it began to feel strong empathy. The same that can generate an actor in a film.
For the moment, therefore, the love story between Samantha, the operating system equipped with general artificial intelligence with the voice of Scarlett Johansson, and Theodore Twombly (played by Joaquin Phoenix), is destined to remain on the big screen of the cinema as told. in 2013 from Her. But if the ability of machines to reason and think independently is still a long way off, we must be aware that the strength and form of artificial intelligence will continue to grow relentlessly.