Rohin Shah

Karma: 4,175

Hi, I’m Rohin Shah! I work as a Research Scientist on the technical AGI safety team at DeepMind. I completed my PhD at the Center for Human-Compatible AI at UC Berkeley, where I worked on building AI systems that can learn to assist a human user, even if they don’t initially know what the user wants.

I’m particularly interested in big picture questions about artificial intelligence. What techniques will we use to build human-level AI systems? How will their deployment affect the world? What can we do to make this deployment go better? I write up summaries and thoughts about recent work tackling these questions in the Alignment Newsletter.

In the past, I ran the EA UC Berkeley and EA at the University of Washington groups.

http://rohinshah.com

Rohin Shah 27 May 2015 1:21 UTC
1 point
0 ∶ 0
on: Effective Altruism Handbook—Now Online
1. I love the choice of content and organization. I’m impressed you were able to keep it fairly short.
2. Initially I was daunted by the size of the handbook (135 pages is a lot for something I expected to be an introduction), but actually it’s not that big—the font is huge. It may be worth it to reduce the font size so that it doesn’t seem as big.
3. The handbook uses ICCIDD for the organization that’s now called IGN.

Rohin Shah 4 Jun 2015 23:19 UTC
1 point
0 ∶ 0
in reply to: Peter McIntyre’s comment on: You Could be the Warren Buffett of Social Investing
Adding on to this question, there are a lot of negative comments on the second article—do you think that represents a vocal minority, or a majority, and why?

It would be interesting to try this at Berkeley as well, although we’d probably have a different target audience depending on where the article gets published.

Rohin Shah 14 Apr 2016 17:10 UTC
1 point
0 ∶ 0
in reply to: David_Moss’s comment on: The Effective Altruism Newsletter & Open Thread – April 2016
Yeah, that’s right, there are currently 9 on the website (http://eaglobal.org/all-eagx-events), though I’m pretty sure there are several more that are being planned.

EAGxBoston and EAGxBerkeley are happening very soon (April 30), while the others are significantly later.

Rohin Shah 14 Apr 2016 17:44 UTC
2 points
0 ∶ 0
in reply to: Andy_Schultz’s comment on: The Effective Altruism Newsletter & Open Thread – April 2016
I don’t know much about taxes—can you explain why this would help? I might want to do it myself :)

Currently though, it seems like you would pay the same amount of taxes in either case. Let’s take a small example—suppose you invested $75 two years ago, which is now worth $100. And let’s suppose it would be $125 in another two years, at which point you would sell it all.

Now currently you have to decide whether to donate $100 in cash, or to sell the $100 of stock and reinvest the $100 of cash into the stock. (I’ve heard of some law that prohibits you from re-buying the exact same stock, to avoid gaming the tax laws, but let’s ignore that for now—in any case, you could buy a different but similar stock.)

If you donate the $100 in cash, then:
- You get a deduction of $100 on your taxes this year because of your donation
- When you sell the $125 in two years, you realize a long-term gain of $50, which you pay taxes on
If you sell the $100 of stocks and reinvest the $100 in cash, then:
- You get a deduction of $100 on your taxes this year because of your donation
- You realize a long-term gain of $25 this year, which you pay taxes on
- When you sell the $125 in two years, you realize a long-term gain of $25, which you pay taxes on
In both cases, you get a deduction of $100 this year, and you pay long-term capital gains taxes on $50 (albeit at different times). So unless you expect that your tax rate will be a lot higher in two years, it doesn’t seem like this is really helping.

In fact, since you sold and re-bought your stocks, you’ve reset the buy date, and now it’s possible that any gains you make will be short-term capital gains (if you sell the stock within a year), which would then be taxed at a higher rate.

Rohin Shah 14 Aug 2016 19:58 UTC
13 points
0 ∶ 0
on: Effective Altruists really love EA: Evidence from EA Global
″… of course, we should strive to be more welcoming. Yet, if EA is already building a highly loyal community willing to promote EA, then perhaps this isn’t one of the most pressing problems facing EA.”

It seems to me that EA is great at getting people who are very much like existing EAs, which leads to the risk of ossification of “types of people in EA” (in other words, a lack of diversity). I think being more welcoming is important to avoid this. (If I remember correctly, you agreed with this point this last year, have you changed your mind on this? If so, how?)

Rohin Shah 16 Aug 2016 22:14 UTC
25 points
0 ∶ 0
in reply to: Kerry_Vaughan’s comment on: Effective Altruists really love EA: Evidence from EA Global
At EA Berkeley, we have members who came through people telling their friends, and people who we got essentially by picking students at Berkeley near-randomly (but selecting for altruistic tendencies). (TL;DR: These students get involved either through a speed Giving Game or by seeing an ad for our course on EA, rather than through social networks.)

The members who came from people telling their friends fit the standard EA demographic, as you might expect—mostly male, and all STEM (and mostly Comp Sci). Let’s call them group A.

Meanwhile, the students who we got near-randomly, selecting for altruistic tendencies, are more likely to be female and have essentially no specific major. Let’s call them group B.

Now, group A forms a much larger percentage of EA Berkeley, so they tend to determine topics of conversation and group activities. We’ve talked about NP-completeness, video games, the popularity of certain CS professors, etc. These are great topics of conversation for group A, even if the individual members don’t know each other. Note that they have nothing to do with EA. However, there are usually one or two members of group B at these events, and inevitably they are quiet and don’t interact much with everyone else.

With that sort of dynamic, it’s not surprising to me at all that we tend not to get very many active members from group B. When I say “being more welcoming”, I mean not talking about NP-completeness, video games and popularity of CS professors, and instead talking about topics that everyone could be a part of. This seems like an important change that will directly lead to more retention of members in group B, which in turn seems important to me because it increases the diversity of EA.

Note that “EA is already building a highly loyal community willing to promote EA” is perfectly compatible with this view—that loyal community is group A.

Also, as a result I strongly disagree with “presumably, similar behaviors work with a wide range of people”—the behavior of “talking about NP-completeness, etc.” works great with CS majors but not others.

I mostly agree with your last paragraph, except that I think “being more welcoming” is the external pressure that would help EA become more diverse (through the mechanism outlined above).

Rohin Shah 17 Aug 2016 8:09 UTC
5 points
0 ∶ 0
in reply to: Kerry_Vaughan’s comment on: Effective Altruists really love EA: Evidence from EA Global
Okay, I agree with the “don’t select on the correlates” phrasing.

That said, when I hear “be welcoming”, and even “be nice”, I don’t hear “Don’t talk about controversial EA-related topics”, I hear something more like “Don’t talk about CS professors”, which I certainly do think we’re failing at. (Heck, we couldn’t do this at EA Berkeley, which already feels more diverse to me than the general EA community.) I don’t know if everyone else means that when they say “be welcoming”.

(Some evidence that other people feel this way too—as of now, 4 people upvoted my previous comment.)

Rohin Shah 18 Aug 2016 22:42 UTC
2 points
0 ∶ 0
in reply to: Ajeya’s comment on: Effective Altruists really love EA: Evidence from EA Global
My current belief is something like “EA is unwelcoming to people not in the standard EA demographic”. So based on that:

Weak evidence:
1. EA demographics have been moving towards “normal” (eg. less gender bias, more racial diversity, but probably still mostly relatively rich people). A priori, I would expect this to happen anyway, but at a pretty gradual rate, something like 1-5 percentage points per year for gender bias.
Moderate evidence:
1. Data that shows that people not in the standard EA demographic are just as likely to be committed EAs after a medium/high interaction that involved meeting other EAs in the real world (eg. physical meetups or EA Global).
2. Data that shows that people not in the standard EA demographic are just as likely to be committed EAs after a low/medium interaction (eg. a Giving Game). A priori, I would expect that they are just as likely to want to learn more, but are less likely to continue on the path after learning more and engaging with the community.
3. Something similar to the above two, like Ajeya’s suggestion.
Strong evidence:
1. The majority of other local EA group leaders disagree with me.
For this issue specifically, I trust observations made by local group organizers more than I would trust large scale observational/correlational data, just because I can imagine so many different factors at play here that even if the data supported the hypothesis that EA as a whole is welcoming, I would still expect there to be several subfactors where we could and should still improve. (Though I could be convinced that it may not be worthwhile to do so.)

Rohin Shah 20 Aug 2016 3:56 UTC
3 points
0 ∶ 0
in reply to: John_Maxwell’s comment on: Effective Altruists really love EA: Evidence from EA Global
Recruit outside your social network: We teach a DeCal (student taught course) about Effective Altruism that is posted on a list of DeCals that all Berkeley students can see, and we play speed Giving Games with random students who are walking down Sproul plaza. The class has been pretty successful; the speed Giving Games not so much (most successes there are with people who already know someone in the club). I’ll be posting a retrospective about EA Berkeley soon with more details. TL;DR: This is hard to do.

I would say “be welcoming” is more important—there’s already a small base rate of people outside standard EA social networks that have some interest in EA. Currently I think a very large proportion of them (>80%) end up not becoming a part of EA. (Compared to ~50% for standard EA demographics.) Bringing that number down would be very helpful, and I think is more tractable.

Re: random unrelated note: That makes sense, but I will say that my impression of “be welcoming” is not “suppress weird ideas”, I’ve said more about this in a comment above.

EAGxBerkeley 2016 Retrospective

Rohin Shah11 Sep 2016 6:27 UTC

18 points

2 comments8 min readEA link

EA Berkeley Spring 2016 Retrospective

Rohin Shah11 Sep 2016 6:37 UTC

6 points

0 comments2 min readEA link

Rohin Shah 22 Sep 2016 19:36 UTC
2 points
0 ∶ 0
on: Review of EA Global 2016 Marketing
Re: Newsletters: I’m on many newsletters and was going to apply anyway (I may have already applied before seeing emails on the newsletters). I would guess that this applies to lots of other attendees (perhaps a hundred or so?). Would that make the newsletters seem less useful? (I’m unclear on your methodology for the mailing list data.)

Also, small note: “The emails stopped if the person clicked a link or if their email address was used in an application.”

I definitely kept getting emails long after I had applied.

Rohin Shah 22 Sep 2016 20:23 UTC
1 point
0 ∶ 0
in reply to: Amy Labenz’s comment on: Review of EA Global 2016
10% of attendees = 100 people

Time spent traveling ~ 20 hours per person (Seems a bit high)

Value of time: $30 per hour (this seems high to me, though I know others claim $50-100. Either way, your conservative estimate shouldn’t be higher than $30, I would think.)

Total: $60,000

It still seems like you’d be saving money according to Ben’s calculation.

(It’s definitely possible though that the time costs would discourage people from attending, resulting in less applications/attendees.)

Rohin Shah 28 Sep 2016 21:37 UTC
1 point
0 ∶ 0
in reply to: Benjamin_Todd’s comment on: If you want to disagree with effective altruism, you need to disagree one of these three claims
But such a critique also falls under the second kind of critique that you said would be a “misfire”. Perhaps you meant that it’s a misfire only if the critic is trying to argue against ideal EA, but in my experience most critics are not trying to do that, they’re arguing against the EA movement.

Rohin Shah 7 Oct 2016 12:15 UTC
0 points
0 ∶ 0
in reply to: Habryka’s comment on: Review of EA Global 2016
Sorry, totally missed the response here.

The question isn’t “How many people have to travel”, but “How many people have to travel who otherwise would not have to”, which basically reduces to “How many people are coming from the Bay Area”. I’ll admit that I don’t really know that number, but it seems implausible that it’s 50%.

(I could believe that it’s 50% for the last EA Global, but I expect that a lot of people showed up because it was so close to them, and so we’d get less of those people but more of other people who live closer to the chosen venue.)

In addition, as Denkenberger said, any participants from the US East Coast or Europe would travel less.

Good point about the heavy-tail distribution for time value, I hadn’t thought about that. I agree that taking that into account $30 may not actually be too high. (Although travel time is not completely wasted, you can work along the way.)

Rohin Shah 12 Oct 2016 19:34 UTC
0 points
0 ∶ 0
on: CEA Updates + August 2016 update
Link to add email address says:

You need permission This form can only be viewed by users in the owner’s organization.

Try contacting the owner of the form if you think this is a mistake. Learn More.

Rohin Shah 19 Oct 2016 5:43 UTC
4 points
0 ∶ 0
on: CEA Update: September 2016
What does Giving What We Can spend its time on? I realized that I am unable to answer this question even loosely, which is unusual. The Our Team page (https://www.givingwhatwecan.org/about-us/team/) helps, but parts are still unclear. More generally, I’m not sure what activities GWWC does that cause more GWWC pledges, which is what it bases its impact on.

What percentage of GWWC pledges do you think were caused by GWWC (or CEA as a whole)? I’m particularly interested in measurements that don’t rely on self-reports. So far, I’ve only seen statistics about self-reports by people who took the pledge, which were used to get a 60:1 multiplier.

Rohin Shah 30 Oct 2016 8:28 UTC
3 points
0 ∶ 0
on: Dedicated Donors May Not Want to Sign the Giving What We Can Pledge
Personally, I think the overjustification effect happens to me a lot, and it would have been better if I had considered that before taking the pledge (though I’m fairly confident I would have taken it anyway). Thanks for writing about that.

However, I’m not too worried about the loss of flexibility, primarily because I don’t see the commitment as that hard. I believe that the pledge is reasonable and lenient in cases like the ones you describe, especially increasing expenditures. I’m not sure where I got this belief from—I think it was something from the Giving What We Can website? In any case, if that’s the norm around the pledge, I don’t think the loss of flexibility should be that important.

Rohin Shah 1 Nov 2016 21:09 UTC
4 points
0 ∶ 0
on: Where I Am Donating in 2016
Re: Open Phil and clean meat—I don’t think the point about researchers is as strong as you imply. Do researchers claim that clean meat will become cost-competitive? It’s possible that their goal is to get fake meat to the point where it was cheap enough that it could be sold to vegetarians, which would still do a lot of good, but not nearly as much.

Basically, I agree with the point that they expect to get useful results and so we should too. But do we know that the useful result in question is “cost-competitive clean meat” and not something weaker? I legitimately don’t know what the answer is, having not researched this in detail. I skimmed the Open Phil report a while ago, and my impression was that researchers didn’t dispute the claim that it would be very difficult to get to the point where it is cost-competitive. (It’s decently likely though that I’m misremembering.)

Rohin Shah 6 Nov 2016 3:02 UTC
0 points
0 ∶ 0
in reply to: aliwoodman’s comment on: CEA Update: September 2016
Thanks!

Side note, is there an easy way to get push notifications, perhaps through email, when there’s a reply to a comment or post you wrote?