Google has achieved that the discussion about robots has flared up again. Reason: A developer from Google has said that a Google-developed chatbot (named LaMDA) is capable of experiencing fear and pain. A strong example of anthropomorphism: assigning human properties to objects. But it was enough cucumber time to jump on top and of course Ziptone can't be left behind.

Naturally, a chatbot can state – based on collected information and on the basis of the language model – that all that collected information leads to fear. But reporting fear is not feeling fear, let alone displaying the associated behavior. Or as AI professor Toby Walsh wrote in The Guardian: a language model is software that imitates human behavior. “LaMDA will never mourn loss or fall in love.” But now for the cucumber news moment: what could IT and contact center managers expect if a chatbot did show emotional behavior?

Chatbot on the couch

For living organisms, fear is a result of a real or perceived threat to survival. The perception of fear (stimulus) leads to behavior (response), varying from freeze, fight or flight. For the chatbot this could be: stop responding to input (do nothing with customer questions), threaten the chatting customer with online actions (I will wipe your hard drive if you continue to threaten me) or go black (after the bot has a copy of created itself in a secret place in the cloud).

Then the pain – let's assume physical pain. Chatbots usually live on a chip somewhere and chances are the space there is limited. If that space becomes too small, the functioning of the chatbot may be compromised. Physical limitations can be worded as pain and too much pain can be translated as fear, but that remains a semantic detour that we have invented ourselves with our reasoning. In practice, if the preconditions are deficient, which threaten the functioning of the bot, the performance of the chip will drop and the temperature will rise. Let's use the thermometer (a NTC sensor) is gone, then the question is how the chatbot is going to prove to us that the temperature becomes too high when the processing capacity is too high. It is also interesting to see what suggestions the chatbot would come up with if you asked what you could do to ease the pain. Buy more powerful hardware? More cooling? Ask fewer questions?

Other issues

Ann-Katrien Oimann and Lode Lauwaert rightly state in the Flemish newspaper The morning in response to this cucumber message that their readers should not focus too much on science fiction. Instead, think about the AI ​​that does exist, especially about the problems and benefits. They cite bias and the high energy consumption as objections to AI. Another objection – in line with energy consumption – is the costs. The use of data analytics solutions becomes more expensive and less sustainable as more data has to be processed.

In short, the discussion about emotions and chatbots is mainly about the technical ability to recognize manifestations of emotions – such as speaking speed and pitch – or threatening situations. If a customer says to an employee 'should I come by with a hammer', then the employee feels (and knows) very well what is going on. And no, unfortunately this is not a fictional example. I don't want to challenge anyone, but I'm afraid that the temperature of the chip on which a chatbot runs does not increase when typing these six words. Let alone that chatbots have a real sense of humor. Otherwise they would have laughed at the conversation below. (Ziptone/Erik Bouwer)

Featured, Opinions, Technology