Patrick Winston is a professor in the computer science department at MIT (http://people.csail.mit.edu/phw/). He heads Genesis, a project dedicated to understanding human intelligence and reproducing it in AI. He specializes in story understanding in artificial intelligence; for instance, he has programmed Genesis to understand the concept of “revenge” in Macbeth.
Winston teaches “6.xxx”, The Human Intelligence Enterprise, every spring. In addition to giving a broad survey of artificial intelligence, he teaches students to communicate effectively in papers and presentations, and tell memorable stories.
One reason humans are so smart is our ability to articulate concepts - and stories - through language. Have you ever had the experience of failing multiple times at something and then “getting it” once someone explains clearly and concisely what you need to do - whether it is throwing a baseball or working an algebra problem? Once you have articulated a problem that you may not have been aware of, or a key principles behind something, the task becomes a lot easier. Here are two examples (from Winston):
A long time ago, Patrick took a painting class. “Look at these trees,” his art teacher said. “The trunks aren’t all brown. They’re shades of gray, black, blue and purple.” Patrick hadn’t noticed this ever before, but now he’ll always see it.
Someone pointed out to Winston that when he wrote recommendation letter, he used the word “charming” for female candidates and not male candidates. “Oh my gosh I do that,” Winston said. He never did it again.
The first edition of his AI textbook was terrible. It was brought to his attention that he didn’t write good introductions to chapters that motivated students. Thereafter he started every chapter with an empowerment promise (see for example http://courses.csail.mit.edu/6.803/chapteropenings.html).
On this page I articulate several key principles behind effective communication (and various other stuff) - tips easy to remember and apply once articulated - as taught by Winston in 6.xxx.
Table of contents
How to be a leader
The beginning of a piece of writing should have an empowerment promise and possibly a hint of entertainment. Convince the reader that (s)he will gain something by reading it. The end of a piece should show the reader that you have delivered on your promise. Tell them “This is what I did” or “This is what you will be able to do now, now that you have read this.” Don’t be afraid to use 1st person.
Include specific details (because details sell!) and use active verbs whenever possible. Don’t expect people to reason too much as they read, so flesh things out. Err on the side of being explicit.
The VSNC paradigm gets your writing read. A good paper should have the following components. All of these should appear in the abstract.
Vision: What is the big idea? What concrete advances will it allow? For instance, “If we are to understand…, then we must…” Be explicit.
Steps: List the steps needed to carry out the idea. Show the reader you don’t just have the idea, you can also do something with the idea.
News: List the results coming from these steps. Be as specific as possible.
Contributions: What did you do that hasn’t been done before? Use “sanctioned verbs” such as prove, demonstrate, implement, test (experiment), frame (problem), survey, identify (key problem), present, speculate, worked out (table, classification).
(Compare to [Understanding by Design|http://www.ascd.org/research-a-topic/understanding-by-design-resources.aspx].)
The VSNC paradigm gets your writing remembered.
Slogan: A repeated phrase that sticks in readers’ minds.
Symbol: A repeated picture that sticks in readers’ minds.
Salient: Have one idea that sticks out. (If you put “too many” good ideas in, then none of them sticks out and the reader won’t know what to make of the piece.) (Warning: this page does not follow this principle.)
Surprise: Something unexpected.
Story: Have a narrative.
Begin each chapter (and begin each lesson in a classroom) with an empowerment promise - what you will be able to do after the lesson - and show that you have delivered upon that promise at the end of the chapter. For an example of a chapter opening, see http://courses.csail.mit.edu/6.803/chapteropenings.html.
Break up into large number of sections/subsections for easy reading. (This limits the amount of time you have to read before you get to a place when you can pause, a kind of “local goal” you can keep aiming towards.) Lots of titles together act as a road map, and point out the salient points of each section.
Making section titles complete sentences helps the reader summarize the main point achieved from reading the section.
Ask people who are not your friends to evaluate your abstract, because they won’t hallucinate details that aren’t there.
Always include your vision, steps, news, and contributions. Having slides with these names is not a bad idea.
Always end with a contributions slide. Never end with a “thank you” or “Questions?” or “Bibliography” slide. When you take questions, people will be looking at the last slide for 10+ minutes, and you want them to see your contributions in those 10 minutes, rather than an empty slide saying “Thank you.”
Don’t say thank you at the end of the talk, especially if you’re female (!). The audience should be thanking you. Instead, salute the audience. “And with this slide I conclude my talk.” “And that concludes what I have to say.” Shake hands with the host. “I really enjoyed being here today.” “You’ve been a wonderful audience. You’ve asked good questions. I look forward to an opportunity to speak to you again.”
Do not use a laser pointer. It’s hard to aim accurately and the audience sees the back of your head. (Put arrows in your powerpoint instead.) Winston, of a laser offender: “We could all have left and he wouldn’t have known.”
People can’t read the slide and pay attention to you at the same time. It overloads the language processing system in the brain. Keep the text on the slides short and punchy, not a copy of (or even more than) what you say out loud.
Make font big.
Always sound excited. If you say “I am excited” then you must sound excited! Use gestures!
How not to make a presentation: (Video where Susan Hockfield and Drew Faust unveils edX.) All the people on the panel look bored. So the audience is going to be bored too.
How to make a presentation: (Steve Jobs unravels the iPhone.) http://www.youtube.com/watch?v=ftf4riVJyqw He sounds excited. He uses the language of a 5-6th grader - that’s part of it too.
Don’t put your hands in your pockets or behind your back. It’s insulting, in several parts of the world. Also, you might have a concealed weapon. If you don’t know what to do, clasping your hands in front of you always works.
During presentations, Winston always has the lights on. “But nobody can see any slides,” the AV person might say. “Nobody can see any slides if their eyes are closed,” Winston says.
A presentation is not a lecture. The main point of a lecture is to teach and the main point of a presentation is to impress.
Winston’s “how to speak” talk: http://www.youtube.com/watch?v=VG9_j6VYaKo
Don’t be a conspicuous job-climber, i.e. don’t keep changing your set of friends. People with the bad kind of ambition keep changing their set of friends. People resent these people, because these people ask you out for lunch just so they can use you for the next step in their plan.
Keep on the radar of several important people who would be useful (10 is a good number). For instance, a friend of Winston had a habit of submitting a monthly piece of news to an art magazine. They didn’t put his submissions in, but he kept at it. One month the writers found they had a gap to fill. “Let’s put this in!” They started noticing this writer, and started to rely on his columns. Eventually they put his column in every month.
The 14 criteria for investment in a business. Idea side
Why now? (Now is a window of opportunity because…)
Channel (for selling the product)
Delivery (can you deliver the whole solution, or do you need to team up?)
Weakness (and a plan to fix it)
Barrier (why do you have an unfair advantage over anyone else who wants to do the same thing?) People side
Integrity (no lying)
Self-confident (but still willing to identify weaknesses and take advice)
Action bias (be action-oriented)
Hire a good team (hire to your own weaknesses. It’s better to hire people who are offscale good at one thing, rather than good at a lot of things)
The difference between a businessman and a normal person: “A businessman makes a decision and builds evidence for why it was the right thing to do. A normal person builds evidence on both sides.”
The biggest reason that people start their own company is to prove an idea to someone else (and especially to prove someone wrong).
When Winston became the head of the AI lab at MIT (CSAIL), he asked 15 other lab directors/department chairs what he needed to do. 14 had no idea. The 15th was Jay Forrester, who built the first computer. They talked for 30 minutes. For the first 25 minutes, Forrester told him why he should not have an AI lab. For 5 minutes, he said, you should be trying to protect the US from air attacks from the USSR.
Moral of the story: You need a mission. “Being the best at X” is not a good motivator. “Being the best AI lab” is not a good motivator. An artificial intelligence lab without a mission is too inhomogeneous and diverse to make any sort of meaningful impact. On the other hand, an AI lab with a single mission can get funds, unite its workers, and make a powerful impact.
You have to make everyone in the organization feel valued. Specifically, everyone has to feel like…
their work is valued,
their work makes a difference,
and they are valued.
If any of these are left out, you get a discontented employee who won’t work hard. How do you convey to people they are valued? Simple: tell them. “Your work is valuable, and makes a difference to me.” “I value your presence in class.” etc. To make sure people feel valued, don’t push everyone in the same hole. Figure out what people like to do, and match people with that work.
Example of no-no: For a time, faculty at MIT graded secretaries (A,B,…) in 10 aspects of work performance. Winston: “This is the dumbest thing I’ve ever seen.” Comparing people linearly like this does nothing to improve performance and breeds resentment. Besides, different people excel in different things.
A person once asked Winston, “I want to start a business but keep the stock to myself. How do I attract employees?” Winston said, “Go find someone else to take advice from.” He didn’t want to talk to anyone who screwed people over.
If you want an example of an organization that breeds loyalty, look no farther than the military. “Are you an ex-marine?” Winston once asked someone. She replied: “There are two types of marines: live marines and dead marines. There are no ex-marines.”
It is important to have principles. It gives something for you to fall back on, when you’re under high pressure and don’t have time to think through all sides of an issue.
(cf. train vs. educate)
Whenever you organize something, make sure you talk to all the people who will be insulted if they are not consulted. Otherwise they will be against your plan from the beginning.
The leader sets the example. Every company has a style; the character of a company is fixed by the top person. (Put example here, on Hockfield at MIT if I can remember.)
Whenever you are meeting someone, look them in the eye. When a committee considers candidates for a position (for instance, the president of MIT), a candidate will have to shake hands with a long row of people. The biggest mistake is to turn your head toward the next person while you are still shaking hands with the previous person. This creates a kind of subconscious bias in the committee member’s heads. These candidates do not get the position.
Make sure you send email to the right person. (Once a student accidentally sent an email to a secretary with the same name as his girlfriend.)
Write emails as if they were going to be printed in the newspaper. Do not send nasty emails, especially on Friday. People have a whole weekend to get fired up.
Rather than send a nasty email, talk face-to-face. Misunderstandings will be resolved much faster. It is easy to kill at a distance.
Do not lie. DO NOT LIE. DO NOT LIE. BECAUSE YOU ALWAYS GET CAUGHT.
Example 1: Winston got a call from an employer, asking about a student.
Winston: I’m sorry, I don’t remember the name. I’ve had many students over the years.
Employer: But this was your PhD student.
Winston: I’m sorry, I really can’t remember. You sure about this?
Employer: But you were his PhD supervisor.
Turned out the student’s resume was entirely fabricated. The most common fabrication on a resume is university degrees. To check this, employers call the university. “Did A graduate?” They will say yes or no, and no more, but that’s all employers need.
Example 2: In 1979, Marilee Jones said on a resume that she had a degree from Rensselaer Polytechnic Institute. She was probably broke at the time, embarrassed that she had nothing more to brag about than having graduated from a local Catholic college. It wouldn’t matter after she got hired. She got hired as a secretary at MIT. She was much loved by all her peers and rose through the ranks to become dean of admissions at MIT, probably the best person to have filled that post. In 1997, it was found that she had never gone to Rensselaer Polytechnic Institute. (When you’re dean of admissions, you make a few enemies.) She was forced to resign from her post, and now has a Wikipedia article as tribute to her lack of integrity (http://en.wikipedia.org/wiki/Marilee_Jones).
Example 4: A Harvard author’s “chic lit” book deal was canceled after an editor found 25 instances of plagiarism. She made the news and also has an everlasting Wikipedia article (http://en.wikipedia.org/wiki/Kaavya_Viswanathan).
Winston catches a few cheaters in his class from time to time. “They are the lucky ones,” he says. They have learned the lesson that you always get caught. The worst thing is to get a reputation and not now why you have it. Oddly, people don’t tell you why if you have a bad reputation.
You don’t learn by asking “How do you do this?” You learn by asking, “Let’s go through an example.” How does a Martian learn how to bag groceries? By going to the supermarket, watching a kid bag groceries, and asking questions when you see 2 things handled differently. “Why are you putting these peas in the bottom of the bag and those on the top?” “Because one is canned and one is frozen.”
Near-miss learning is learning from knowing which examples work, and which examples almost work (so you can say exactly why they fail to work). This is an important concept in both human learning and AI learning. A computer program that is fed examples in a specific order following the philosophy of near-miss learning will learn much faster.
How to do good research
Devise representation (Make the right stuff explicit, expose constraints, make it computable.)
Do not fix on an algorithm first, then find a problem that uses the algorithm. (Problem with AI research - people go to conferences on “Bayesian nets” - concentrating on an algorithm rather than a problem.)
How to write a good academic reference letter: Say why the person’s research matters to you, personally. Not why the person’s research is important, in general.
The trouble with the tenure process: The Trouble With Physics: The Rise of String Theory, The Fall of a Science, and What Comes Next, Lee Smolin
‘’Keep in mind what makes people’s lives easier.’’ Once Winston was in charge of organizing an AI symposium. This was in the 90’s, after the AI bubble crash in the 80’s - when AI was too focused on so-called “expert systems.” Naturally business wasn’t very enthusiastic about AI at the moment. In fact, when businesspeople came across bad ideas, they would say “It’s just another AI.” Yes, it was that bad. So Winston organized a symposium to turn businesspeople’s heads towards AI again. Winston briefed the speakers about how to make their talks appealing to business. The talks went on, one after the other, but still, no one seemed to be very enthusiastic.
On the last night of the conference, Winston thought long and hard about what to do. He decided to make a presentation at the end, summarizing the take-away for business about AI research. Essentially, it was a huge “contributions” section. He called the presentation, “Thinking about rethinking AI.” To make sure they got the take-away message, he made sure all participants had copies of the transparencies, and gave them a web address where they could download the slides.
In a survey, the response to the symposium was overwhelmingly positive - it pretty much got the highest marks for any symposium. Why? Winston himself didn’t quite get it until some time after.
Think it through from the businessman’s perspective. Boss says, go to this conference, find out what this AI crap is about. Oh, and write me a trip report. Businessman goes, unwillingly, trudges through all these long presentations. At the end, Winston makes a presentation and basically delivers a trip report into his hand. All he has to do is to download the slides - on which Winston wrote the contributions of AI to business - and put a different frame about it. His message gets through to all these companies.
’’It’s easy to overestimate what you can do in a year, and underestimate what you can do in ten.**
Rumpelstiltskin principle: When you give a name to something, you gain power over it.
Reification: Turn an abstract idea into a something concrete (ex. a cell in a data model).
Goldilocks principle: To classify things, look at their **intermediate’’ features, not too small and not too large.
Winston’s theories of human intelligence.
Inner language hypothesis
Strong story hypothesis
Directed perception hypothesis
Social animal hypothesis
Exotic engineering hypothesis
Get the message to Garcia! Take the initiative to do what’s needed to get things done, rather than rely on other people to tell you all the steps you’re supposed to take. http://courses.csail.mit.edu/6.803/pdf/hubbard1899.pdf
People are remembered for 1 thing. It’s easier to latch onto one project to do really well, then try to do a lot of different ones.
There are two types of secrets: type 1 secrets and type 2 secrets. Type 2 secrets are the type you tell one person at a time.
‘’How to know what’s really going on on a college campus’’: Read the school newspaper.
“Don’t be too proud to slap.” (In wrestling, if you fall, you have to slap the mat so you don’t get injured.)
“Bark like a dog to warm up your voice.”
Give reasons, because people can’t argue with reasons (as easily). Don’t say “Sorry, you didn’t make the basketball team.” Say, “You didn’t make the basketball team because BLAH.”
Difference between illustration and demonstration: An illustration is 1 example. A demonstration is many examples.
One Fields medalist writes the names of concepts on index cards. To decide what to research next, he randomly takes two and thinks, Is the overlap between these concepts interesting? (I don’t know the truth of this story.)