What 101 New Yorkers said at their door

Question 1 · Sentiment · n=101

When it comes to artificial intelligence, are you more excited, more concerned, or about equal?

• More excited18 of 101
• About equal37 of 101
• More concerned46 of 101

Question 2 · AI Usage · n=101

How often do you use AI — daily, weekly, monthly, or never?

• Daily33 of 101
• Weekly22 of 101
• Monthly12 of 101
• Never34 of 101
• Not sure0 of 101

Question 3 · Development Preference · n=101

Should superhuman AI be developed, or never developed?

• Should be developed38 of 101
  • As fast as possible1 of 101
  • With current regulations1 of 101
  • With strict safety rules11 of 101
  • Only after proven safe25 of 101
• Never58 of 101
• Don't know5 of 101

Question 4 · Risk of Loss of Control · n=101

How likely is it that superhuman AI could become too powerful for humans to control?

• Very likely55 of 101
• Somewhat likely25 of 101
• Not very likely12 of 101
• Not at all likely4 of 101
• Don't know5 of 101

Question 5 · International Treaty · n=101

Should countries sign an international treaty — like the ones for nuclear weapons — to limit how powerful AI can become?

We asked this with the words “like the ones for nuclear weapons.” That comparison probably pushed support up.

83 of 101 supported in total

• Strongly support60 of 101
• Somewhat support23 of 101
• Somewhat oppose4 of 101
• Strongly oppose4 of 101
• Don't know10 of 101

Question 6 · Development Pace · n=101

Do you think AI is being developed too fast, too slow, or at about the right speed?

• Too fast59 of 101
• About right23 of 101
• Too slow1 of 101
• Don't know18 of 101

Question 7 · Voting on AI Limits · n=101

If a candidate promised strict limits on developing more powerful AI — would you be more or less likely to vote for them?

We changed one word in this question four days into the fieldwork. The 44 people answered before that day heard “strict controls”; the 52 answered after it heard “strict limits.” The 5 answered on the day of the change could be either — we cannot tell which. We changed the word because “controls” was being heard as the candidate wanting to have control, rather than to impose limits, which is close to the opposite of what we meant. Nothing here separates the two groups, so this is the weakest answer on the page: read it as a rough sense of direction and not as a measurement.

• Much more likely23 of 101
• Somewhat more likely28 of 101
• No difference32 of 101
• Somewhat less likely7 of 101
• Much less likely5 of 101
• Don't know6 of 101

Question 8 · Timeline · n=101

How soon do you think AI could become far smarter than any human?

• Already has20 of 101
• Within 5 years40 of 101
• 5 to 20 years21 of 101
• More than 20 years8 of 101
• Never9 of 101
• Don't know3 of 101