SUNY Geneseo Department of Mathematics
Monday, November 8
INTD 105 17
Fall 2021
Prof. Doug Baldwin
(No.)
The next essay asks you to “join a conversation” about Alan Turing.
See the handout for details.
Key dates:
Don’t forget that you need to meet with me to discuss the final version of your Enigma essay, too.
If you haven’t met with me on past essays (Warm-Up and Race, each with one meeting about the draft and one about the final version), now is the time to start catching up. The semester ends sooner than it might seem, and there won’t be time to finish everything at the last minute.
The branch of computer science that deals with trying to make computer programs behave like humans or other intelligent beings.
Whether it’s possible for a computer to be intelligent (which is a higher standard that just acting intelligent) was a question that fascinated Turing.
He proposed the Turing test as a way to determine whether a computer is intelligent or not. The idea is that a person communicates via typing with a partner who they can’t see, so they have no physical cues about whether that partner is a person or a computer. After talking about anything they want, for however long they want, the person guesses whether their partner was a person or a computer. If they are no better at telling the difference than if they just picked a guess randomly, then the computer is said to be intelligent; if it is distinguishable from the human then it isn’t.
The following are some attempts at chatbots that can pass for intelligent:
(Most of these are meant for recreational use, except possibly the first. The companies that run them generally promise not to share the information you give their chatbots, but it’s still probably wise to not to share anything with them you wouldn’t share with any stranger. Chatbots are generally trained on real human conversations, and so may say offensive things that appeared in those conversations.)
Form groups of 2 or 3 people. Each group split into 2; and conduct “conversations” via text message or some other person-to-person text channel between the two subgroups as follows:
The goal is for the first subgroup to decide, after a few turns of the conversation, whether they were talking to a chatbot or to a person. Once they’ve decided, let them know whether they were right, and start a new conversation.
People found it pretty easy to tell when they were talking to a chatbot. Often the ’bots would give generic or vague answers to questions where a person would be more specific.
In fairness to chatbots, they work better when the subject of the conversation is restricted. Thus the chatbots that show up on commercial websites (e.g,. banks, travel agents, etc.) to help visitors can actually be pretty good at offering advice in their narrow domain.
Almost everyone tried person-to-chatbot conversations. Only one group seems to have included a person-to-person conversation, and in that conversation the person was trying to imitate a ’bot. In another group, the chatbot’s statements were edited slightly by a person to make them sound more natural, but the partner still correctly identified them as coming from a program, not from a person.
So at the moment, the kinds of chatbot that are available recreationally on the Web don’t quite seem up to Turing’s standard of intelligence.
Turing as code-breaker, some of his ideas for attacking Enigma.