Machines still have a long way to go before they learn like humans do – and that is a potential danger to privacy safety and more.
By: Aviva Hope Rutkin
Published: BBC, 11 April 2017, http://www.bbc.com/future/story/20170410-how-to-fool-artificial-intelligence
Level of Difficulty: **
BEFORE YOU READ
· How safe is AI? https://www.voanews.com/a/how-safe-artificial-intelligence/2915474.html
· The Truth Behind Artificial Intelligence | Andrew Zeitler | TEDxStMaryCSSchool
· Understanding Artificial Intelligence and Its Future | Neil Nie | TEDxDeerfield
1. Read the example of the self driving car. What conclusion can be drawn from this example? Where else could this conclusion be placed in the text?
2. Look back at the first four paragraphs and match the vocabulary items with their meanings. There are more choices than you need:
The vocabulary items: unbeknownst, far-fetched, adversarial, bamboozle
Meanings: unseen, with no knowledge of, opposing, ignorant, hard to believe, distant, hostile, attack, confuse
3. AI is not as fool-proof as some may mistakenly assume due to the fact that…
4. Read the description of human and AI learning to the end. Why might a machine confuse a panda with a gibbon or a school bus with an ostrich?(Two possible answers, find both) The source of the problem is the fact that…
5. Now look back at the section you have just read starting “It’s something that is a growing concern..” and ending “Someone likely already has”. What subtitle would you give this section:
AI Can’t Think, The Downfall Of AI, The Loop Hole, Algorithms And The Brain.
6. Access the following, scroll down and watch the video: Deep neural networks are easily fooled http://www.evolvingai.org/fooling. The basic reason why AI can make what seems to us, as bizarre mistakes is that AI sees ………………………………..
7. The example of spam filters and how spammers get round them is provided to support the contention that…
8. Look at the paragraphs beginning “What might this allow…” and the following paragraph. Where would you place the following:
· There are various ways in which hackers could cause problems for AI
· There are various possibilities that come to mind
9. Now look back at the three paragraphs concerning adversarial attacks and the quote from Voyobeychnik. What subtitle would you give this section; mark the odd one out:
· The possibilities are endless
· Think it is safe? Think again
· They can and they will
· Tweak and enter
10. Read the example of the glasses. This example is provided to show that adversarial attacks could be ……………
11. Read on until the end of the text starting with “In the meantime…”. Two ways to improve algorithms would be to adopt a ……………………… policy and make programs …………… so that they can withstand an attack.
12. What reminder is provided in the last section of the text concerning the role of AI in our lives?
Write an essay in which you discuss to what extent we will be able to depend on AI in the future.
THE TINY CHANGES THAT CAN CAUSE AI TO FAIL; KEY AND TEACHERS’ NOTES
AI is agreed to be a major development to be reckoned with in the future but will they be “the perfect oracles of truth” or will humans always need to be in the loop? The text discusses this issue and more by providing irrefutable technical reasons. It is a very topical issue and should lead to plenty of discussion.
1. Artificial intelligence can be fooled
2. Unbeknownst: with no knowledge of
Far-fetched: hard to believe
3. Algorithms are being used more and more
4. Because the computer doesn’t pay attention to high level details / Because a computer doesn’t consider the whole picture ; The computer looks at the individual pixels of the picture
5. The Loop Hole
6. Psychedelic images of abstract patterns and colors (that look like nothing much to humans)
7. Vulnerabilities exist and someone will figure out how to exploit them.
8. At the beginning of the second paragraph, After the first sentence in the second paragraph
9. The last one
11. A “the tougher, the better” policy / more resilient
12. Machine learning systems are a tool to do reasoning