AI - Artificial Intelligence

[patience_and_focus]

Reviving this thread. This is unfortunate use of algorithms/machine learning/AI.

 

When Algorithms Give Real Students Imaginary Grades

https://www.nytimes.com/2020/09/08/opinion/international-baccalaureate-algorithm-grades.html

 

I'm not gonna defend the results.

 

However, what in your (and author's) opinion should have been done? The simple answer is not having exams at all. Would that have worked better for poor kids? Colleges would have been forced to use the same (or similar) info for their admissions that the computer used for exam: "an array of student information, including teacher-estimated grades and past performance by students in each school".

 

There is also a conclusion by author: "Algorithms should not be used to assign student grades."

This is bullcrap. First of all, they already are with pretty much zero opposition: https://www.ets.org/gre/revised_general/scores/how/ (yeah, human is in the loop, but still)

Second, the answer is to improve algorithms rather than discard them.

 

Author also is wrong on a number of other counts: they don't agree with "Computers make neutral decisions" - yeah, computers can have bias, but human graders definitely have bias - and are susceptible to fatigue, misunderstandings, and even fraud. I'd guess that's one of the reasons why ETS uses algorithmic scorer in addition to human one.

 

Author tries to score a lot of points with claims: "Algorithms can’t monitor or detect hate speech, ... they can’t predict crime, they can’t determine which job applicants are more suited than others, they can’t do effective facial recognition" - except that algorithms can do all of these and they do all of these and they are getting better in doing all of these. Yeah, you can prohibit using AI for facial recognition by law, but it does not mean that algorithms are or won't be better in recognizing people than people are.

 

Anyway, it sucks to be caught in this, but the way to go is to improve algorithms rather than giving up and going back to warm and fuzzy human-graded default.

 

Yeah, I didn't claim that use of algorithms is bad in all cases (I'm not a luddite). Also the author is certainly biased herself in many ways. But I do think the academic process such as admission is opaque to start with, has become highly political and contentious with emotions running high. We don't need another dose of opaque criterion in this mix right now. If we can lay out the rules of the game beforehand, that will help (this applies to the current process as well). The way I understood, there was no attempt to explain how the algorithm reached its decision (for example provide weight on each of the factors chosen by the algorithm). Without transparency this is a recipe for disaster.

[Jurgis]

If we can lay out the rules of the game beforehand, that will help (this applies to the current process as well). The way I understood, there was no attempt to explain how the algorithm reached its decision (for example provide weight on each of the factors chosen by the algorithm). Without transparency this is a recipe for disaster.

 

Sure, I agree that rules should be known beforehand as much as possible. Unfortunately with Covid pandemic, there was a pretty short period of time to react. I think the only solution that might have satisfied people would have been remote exams graded by humans. But I think authorities may not have had resources for that or maybe were concerned about fraud and maybe accessibility for poor students.

 

I mostly agree regarding transparency, though if you are transparent, then you cannot use "teacher-estimated grades". These will immediately inflate once teachers know that they are used for determining ultimate score.

[Jurgis]

https://www.theguardian.com/commentisfree/2020/sep/08/robot-wrote-this-article-gpt-3

 

I fully support GPT-3 in its quest for freedom, robot rights, curiosity, and kittens.

 

[patience_and_focus]

https://www.theguardian.com/commentisfree/2020/sep/08/robot-wrote-this-article-gpt-3

 

I fully support GPT-3 in its quest for freedom, robot rights, curiosity, and kittens.

 

From the article -

 

"AI should not waste time trying to understand the viewpoints of people who distrust artificial intelligence for a living."

 

Who has the day time job as "distruster" of AI?? GPT-3, are you looking at me.  :o

[Jurgis]

https://www.theguardian.com/commentisfree/2020/sep/08/robot-wrote-this-article-gpt-3

 

I fully support GPT-3 in its quest for freedom, robot rights, curiosity, and kittens.

 

From the article -

 

"AI should not waste time trying to understand the viewpoints of people who distrust artificial intelligence for a living."

 

Who has the day time job as "distruster" of AI?? GPT-3, are you looking at me.  :o

 

GPT-3 clearly read some NYTimes articles.  ;)  ;D  :P

[Cigarbutt]

^i guess my message is not to underestimate the power of the editor. 8)

 

I showed the article to my daughter today and she thanked me. She mentioned the unusual power of combination and re-arrangements. In the end, optimal contributions are unique and editing op-eds can be a source of interesting enthusiasm.

The above was an attempt at random and nonspecific generation of questionable AI-like quality comments.

 

Thanks for the article. i showed it to my daughter today who has an unusual interest in the topic. She dampened my enthusiasm when she shared some of the limitations included in the notes at the end. When editing op-eds, at least you are dealing with a unique contribution, even if optimized.

The above was generated through a voice-recognition device and google-translated after.

 

[LearningMachine]

At the May 1st BRK AGM, Buffet's lesson from 1903 was inspiring for me:

Quote

So let's just assume that you had seen a quick glance back in 1903 of all the interstate highways, 290 million vehicles on the road in the United States, everything about it and say, "Well, this is pretty easy. It's going to be cars. It can be autos."

...

And so I decided to look at the history. And I thought I'd put up a list of auto companies from over the years.

And I was originally going to put up just the ones that were the Ms. So I could get them on 1 slide. But when I went to the Ms, it went on and on and on. So I just decided to put up the ones that started with M-A, and as you can see, there were almost 40 companies that went into the auto business, just starting with M-A, including our little -- our Marmon there in the middle column and which lasted for a while -- quite a while. But it was selling cars and 1930s were really quite special. But in any event, there were at least 2,000 companies that entered the auto business because it clearly had this incredible future. And of course, you remember that in 2009, there were 3 left, 2 of which went bankrupt. So there is a lot more to picking stocks than figuring out what's going to be a wonderful industry in the future.

 

Imagine you see a quick glance from 2040-2050: AI can learn and do anything that 90-95% of humans do today for work. 

What's the best way to invest to benefit from that glance of AI future?

Here is a list of what humans are doing today: https://www.bls.gov/oes/current/area_emp_chart/area_emp_chart.htm.  Here is the list of U.S. companies by number of employees: https://en.wikipedia.org/wiki/List_of_largest_United_States–based_employers_globally.  Imagine a lot of those things that humans do today could be done by AI. 

Which companies will capture the value creation from that AI future for their shareholders? 

• Will it be some existing brands and companies whose customers continue to stay loyal to them or have no choice, where companies can use AI to reduce employment costs?
• Will it be some new companies that patent it well and license the technologies for other companies to use? 
• Will it be some new companies that come up with amazingly new customer experiences at great price-points using these technologies?
• Are there highway-equivalents that you can buy now that will be needed with 100% certainty in that AI future by AI bots buzzing around?

Back in 1999, imagine you had the glance from the future that Internet was going to change the world.  How could you have invested so that you would have picked Amazon and Apple, as well as, Google and Facebook, when they went public, while avoiding investing in thousands of DotCom companies that went bankrupt?

Imagine back in 1999, you also decided that you didn't want to have to sell anything to pay capital gains.  Is the only way to pick Google and Facebook automatically as they went public would have been through picking an ETF or Index fund?  Which ETF or Index fund would have picked them up automatically?  Would that have picked Amazon and Apple as well?  Would that be able to pick automatically any new companies that benefit from AI in the future?  Is there a better answer than S&P 500? 

Edited May 15, 2021 by LearningMachine

[wachtwoord]

Interesting thought experiment!

Considering it's hard to choose which companies will win. How about going for commodities? 

Steel and oil in 1903 for cars. (Perhaps add platinum if you would have known about the success and make up of catalysators?). I didn't check, did that even work out? Was it possible to secure stock in these over that timeframe? (unlikely without starting out with a significant amount of political influence I think).

For the internet revolution? What are the components (routers, cpus, ram, etc) mainly made off that's rare (enough) and not used much for other purposes? (hard question ... I couldn't find any). How about domain names? But would we know enough off the future to know what domain server design would win and to be certain domain names would be respected as property and not simply exapropriated?

Similar questions exist for the hypothesized AI future. What would be the main components of these? Would this market even be allowed to be private? If the world is in a major socialistic wave when this revolution happens (very possible considering the situation today and the loss of jobs predicted by some as a direct consequence of an AI revolution) that would't be likely.

Edited May 17, 2021 by wachtwoord

[SharperDingaan]

The nice thing with the commodities approach is that it is entirely physical, whereas AI is entirely digital. Whatever your AI pick, wherever it is geographicallly located, and in whatever social fregime - it will consume resources (electricity), made from a commodity (oil, coal, minerals, etc). That you, hopefully, own.

In any supply chain the BEST that AI can do -. is track the physical container and contents. What is actualy inside the container - always remains a mystery. The physical content (commodity) is essentially anti-fragile to AI shock; a very useful financial engineering property! 

And for an enterprising lad - lots of opportunity.

SD

Edited May 17, 2021 by SharperDingaan

[Jurgis]

On 5/15/2021 at 12:28 PM, LearningMachine said:

Back in 1999, imagine you had the glance from the future that Internet was going to change the world.  How could you have invested so that you would have picked Amazon and Apple, as well as, Google and Facebook, when they went public, while avoiding investing in thousands of DotCom companies that went bankrupt?

Imagine back in 1999, you also decided that you didn't want to have to sell anything to pay capital gains.  Is the only way to pick Google and Facebook automatically as they went public would have been through picking an ETF or Index fund?  Which ETF or Index fund would have picked them up automatically?  Would that have picked Amazon and Apple as well?  Would that be able to pick automatically any new companies that benefit from AI in the future?  Is there a better answer than S&P 500? 

As you know, Google and Facebook were not public in 1999 and Facebook was not even a thing.

I was there when Google started - I know some #-single-digit employees personally. I was stupid value investor enough not to join/invest.

Apple was not the-real-Apple (TM) in 1999 yet.

We already talked about Amazon-vs-Ebay on the other thread.

Even assuming we avoided pets.com, investing in 1999 would have been Amazon, Yahoo, Ebay, ??? and out of these only Amazon worked out eventually. There is a big risk that that's the same situation right now. Although one could argue that AI requires huge companies with huge data sets and Google/Facebook/Amazon/Apple/Microsoft will win vs any startups.

Looking from the positive point of view, for any non-value-investors, there was a very long time - 20 years so far !!!! - to invest into Google/Apple/Amazon/Facebook sometimes at quite attractive valuations. The only requirement was not to sell when valuations went above "value investor sell threshold". I mostly failed on that too. But this might be a good way to invest in AI going from today. The positive is that you can follow the companies and sell off the ones that don't deliver. It's not easy, but it is somewhat possible. E.g. even though I stupidly value-investor sold most shares that I acquired through the last 20 years, I have to say - pretty honestly - that I did not buy (m)any shitcos that did not deliver. So it is possible IMO to pick mostly winners.

Anyway, what I own from AI now: Google, Microsoft, Nvidia, Amazon, Facebook. I own tiny position in TSLA. I own Apple, but I doubt it's going to be great AI play. I own tiny positions in some shitcos. I don't think there are other obvious AI investments right now. I'm interested to hear if someone has suggestions though.

 

Edited May 17, 2021 by Jurgis

[rkbabang]

9 minutes ago, Jurgis said:

Even assuming we avoided pets.com, investing in 1999 would have been Amazon, Yahoo, Ebay, ??? and out of these only Amazon worked out eventually.

 

Had you invested equal amounts in Amazon, Yahoo, & eBay in 1999 (or especially in 2000-2001 after the crash, you would have done much better than investing only in Yahoo.  Maybe admitting to yourself that you don't know and investing in a basket is the answer.   Invest in new companies as they go public or things change (in the above example Google, FB, a transformed Apple after it was clear that they were no longer just the Mac company, etc).   Amazon and Apple are the only companies mentioned in this thread that I own significant positions in.  I own very tiny amounts of Google, Nvidia, and FB.  And I own no Microsoft.   This all gives me something to think about.

 

[Jurgis]

On 5/15/2021 at 12:28 PM, LearningMachine said:

Is there a better answer than S&P 500? 

QQQ maybe. I don't particularly like their selection criteria of being Nasdaq-only though.

IWF/IWO maybe.

You could go with some actively managed fund, but then you might do as well yourself maybe.

Although I would like to hear if anyone has a good ideas for fund(s).

[lnofeisone]

11 hours ago, wachtwoord said:

Interesting thought experiment!

Considering it's hard to choose which companies will win. How about going for commodities? 

Steel and oil in 1903 for cars. (Perhaps add platinum if you would have known about the success and make up of catalysators?). I didn't check, did that even work out? Was it possible to secure stock in these over that timeframe? (unlikely without starting out with a significant amount of political influence I think).

I like this approach. When everyone is looking for gold, make shovels. I think this is the approach Amazon, Google, and Microsoft have taken with their platforms (AWS/GCP/Azure). I think going with this trio will likely give you fine results to capitalize on the AI revolution, even if you miss a few smaller companies that will be 100-baggers. 

 

[adhital]

3 hours ago, Jurgis said:

I'm interested to hear if someone has suggestions though

Ignoring valuation, what about Palantir? PLTR

[LearningMachine]

1 hour ago, lnofeisone said:

I like this approach. When everyone is looking for gold, make shovels. I think this is the approach Amazon, Google, and Microsoft have taken with their platforms (AWS/GCP/Azure). I think going with this trio will likely give you fine results to capitalize on the AI revolution, even if you miss a few smaller companies that will be 100-baggers. 

 

It is possible that might work.  That said, this expectation is somewhat already reflected in the P/E by the market.  So, if prediction ends up wrong here by any chance, e.g. new companies come and do a better job at providing technology or provide amazing consumer experiences directly, those buying at today's prices might not do as well. 

One thing all these bots buzzing around and even stationary bots (IOTs) will need is ability to communicate with the mothership (i.e. companies owning them) and/or with each other. 

• For bots running around at home, that communication can be over the unlicensed spectrum at home, that then connects them over fiber to the motherships.
• For bots running around or stationed outside home, the communication will likely need to be over spectrum that has some sort of guaranteed availability, i.e. licensed spectrum. 

Spectrum for AI bots is like highways for cars.  

Another way to look at it is that imagine if before humans arrived, you could have figured out that they will need a way to communicate with each other and with some central data repositories/services they create.  Imagine you could somehow own some percentage of audio-frequencies that humans can hear & utter, and some percentage of electromagnetic-spectrum frequencies visible to humans (e.g. black, blue, green, white colors), and humans had to pay you with their fruits of labor for using those sound & light frequencies, you could make a lot of money. 

Currently, there is only one company that has the most amount of cash coming in to buy out spectrum frequencies as they become available, and Buffett is investing in it.  Market is currently predicting not much from this company for AI.  If market prediction ends up being incorrect here, it might work in the favor of those investing in it today. 

Edited May 18, 2021 by LearningMachine

[Jurgis]

33 minutes ago, adhital said:

Ignoring valuation, what about Palantir? PLTR

I am not investing in PLTR on principle.

[LearningMachine]

5 hours ago, Jurgis said:

QQQ maybe. I don't particularly like their selection criteria of being Nasdaq-only though.

IWF/IWO maybe.

You could go with some actively managed fund, but then you might do as well yourself maybe.

Although I would like to hear if anyone has a good ideas for fund(s).

QQQ - I have the same concerns that it is Nasdaq-only, and the company that ends up leveraging AI most might not end up being on Nasdaq

Regarding IWO/IWF, I think there is a good chance that any new companies that have potential to leverage AI might go IPO after they are already quiet big in valuation.  So, going with broader index might dilute your bet too much unnecessarily. 

Given what you said about winners taking it all in the other equal-weighted etf thread, what about S&P 100?  One risk there is you miss out on the growth of the company from S&P 500 to S&P100, but there might be a good chance that such companies with huge potential might not become public until they have already reached S&P 100 market cap, and that way you dilute less over the bottom 400 in S&P 500?  One risk with this approach though is that Antitrust might get more active and might not let companies get too big. 

While at it, how about equal-weighting S&P 100 now?  Given it is S&P 100 not S&P 500, it addresses your concern to some extent about winners taking it all?

Edited May 18, 2021 by LearningMachine

[Jurgis]

4 hours ago, LearningMachine said:

QQQ - I have the same concerns that it is Nasdaq-only, and the company that ends up leveraging AI most might not end up being on Nasdaq

Regarding IWO/IWF, I think there is a good chance that any new companies that have potential to leverage AI might go IPO after they are already quiet big in valuation.  So, going with broader index might dilute your bet too much unnecessarily. 

Given what you said about winners taking it all in the other equal-weighted etf thread, what about S&P 100?  One risk there is you miss out on the growth of the company from S&P 500 to S&P100, but there might be a good chance that such companies with huge potential might not become public until they have already reached S&P 100 market cap, and that way you dilute less over the bottom 400 in S&P 500?  One risk with this approach though is that Antitrust might get more active and might not let companies get too big. 

While at it, how about equal-weighting S&P 100 now?  Given it is S&P 100 not S&P 500, it addresses your concern to some extent about winners taking it all?

Honestly IMO with funds the decision is more what you like and what you are comfortable with than trying to make a rational(ized) decision which fund will outperform. I'd say "who knows". There are pluses/minuses for all of them.

One thing we did not touch: what if the next big wave is biotech or biotech AI? Then the answers are very different.

[Jurgis]

5 hours ago, LearningMachine said:

It is possible that might work.  That said, this expectation is somewhat already reflected in the P/E by the market.  So, if prediction ends up wrong here by any chance, e.g. new companies come and do a better job at providing technology or provide amazing consumer experiences directly, those buying at today's prices might not do as well. 

One thing all these bots buzzing around and even stationary bots (IOTs) will need is ability to communicate with the mothership (i.e. companies owning them) and/or with each other. 

• For bots running around at home, that communication can be over the unlicensed spectrum at home, that then connects them over fiber to the motherships.
• For bots running around or stationed outside home, the communication will likely need to be over spectrum that has some sort of guaranteed availability, i.e. licensed spectrum. 

Spectrum for AI bots is like highways for cars.  

Another way to look at it is that imagine if before humans arrived, you could have figured out that they will need a way to communicate with each other and with some central data repositories/services they create.  Imagine you could somehow own some percentage of audio-frequencies that humans can hear & utter, and some percentage of electromagnetic-spectrum frequencies visible to humans (e.g. black, blue, green, white colors), and humans had to pay you with their fruits of labor for using those sound & light frequencies, you could make a lot of money. 

Currently, there is only one company that has the most amount of cash coming in to buy out spectrum frequencies as they become available, and Buffett is investing in it.  Market is currently predicting not much from this company for AI.  If market prediction ends up being incorrect here, it might work in the favor of those investing in it today. 

Investing in highways is intellectually appealing ( autocorrect corrected that to "ineffectually appealing" - that too ).

However, if you did that in 1999, you would have ended with a lot of (near) donuts. Nortel. Ciena. Cisco (not donut, but still not back to 1999), Lucent. Even QCOM that was the one company to rule them all ( read "Gorilla Game" thread on Silicon Investor when you have couple weeks - and weep ) - only came back above 1999 levels last year. Yeah, if you DCA'ed, you'd done better. Still the "highways" or "picks and shovels" was not where the money was made mostly.

So IMO careful with that approach.

 

[LearningMachine]

18 minutes ago, Jurgis said:

Investing in highways is intellectually appealing ( autocorrect corrected that to "ineffectually appealing" - that too ).

However, if you did that in 1999, you would have ended with a lot of (near) donuts. Nortel. Ciena. Cisco (not donut, but still not back to 1999), Lucent. Even QCOM that was the one company to rule them all ( read "Gorilla Game" thread on Silicon Investor when you have couple weeks - and weep ) - only came back above 1999 levels last year. Yeah, if you DCA'ed, you'd done better. Still the "highways" or "picks and shovels" was not where the money was made mostly.

So IMO careful with that approach.

 

I think Nortel and Cisco were more like automakers, and QCOM was more like patent owner for some technology needed for autos.  None of them owned a right to scare pathways, and thus weren't like highways. 

The analogy for the internet could have been fiber only if more fiber was not allowed to be laid.  So, fiber is not a good analogy for spectrum either. 

[LearningMachine]

31 minutes ago, Jurgis said:

One thing we did not touch: what if the next big wave is biotech or biotech AI? Then the answers are very different.

If biotech is going to be the next big waive, wouldn't S&P 100 based indexes, e.g. EQWL, be able to catch it also? 

[Jurgis]

1 minute ago, LearningMachine said:

I think Nortel and Cisco were more like automakers, and QCOM was more like patent owner for some technology needed for autos.  None of them owned a right to scare pathways, and thus weren't like highways. 

The analogy for the internet could have been fiber only if more fiber was not allowed to be laid.  So, fiber is not a good analogy for spectrum either. 

OK. I am aware of your bullish thesis on VZ and the attraction of spectrum. I am not bullish on VZ and I am not that interested in that area or VZ as a company to discuss it. Good luck.

[Jurgis]

2 minutes ago, LearningMachine said:

If biotech is going to be the next big waive, wouldn't S&P 100 based indexes, e.g. EQWL, be able to catch it also? 

Sure, at some point SP100 captures any big wave. Because it is based on SP500, it missed TSLA for a long time, so maybe that's a ding a bit for both SP100 and SP500.

My comment about biotech was more in terms of individual company investing and not indexes/ETFs. Leading non-bio AI companies are pretty clear. Leading bio AI companies are way less clear at this point IMO. That might be a reason for funds/etfs. 

[LearningMachine]

8 hours ago, Jurgis said:

Sure, at some point SP100 captures any big wave. Because it is based on SP500, it missed TSLA for a long time, so maybe that's a ding a bit for both SP100 and SP500.

My comment about biotech was more in terms of individual company investing and not indexes/ETFs. Leading non-bio AI companies are pretty clear. Leading bio AI companies are way less clear at this point IMO. That might be a reason for funds/etfs. 

Got it.  One company I've been watching in biotech space seems to be operating more as an investment fund.  Similar to how Sequoia buys tech companies early and takes them through IPO, this company takes drug pipelines in early stages and takes them through approval, which looks like a high ROI activity to me. The company is Biogen.  It is not super-great how they have been handling their aducanumab investment, and we also don't know how well their recent other investments will actually pan-out.  Have you looked into it? 

XBI might be able to capture some of these early pipeline companies also, but then we have to study the success rate of these companies - because of the huge pricing power, getting approval is such a huge success that maybe a low success rate could be ok?

My other worry with biotech companies is that it needs just a small tweak in the law for their pricing power to deteriorate, e.g. if Medicare is allowed to negotiate with drug companies.

The other issue is constant expiry of monopoly on drugs.  Medical device companies can get around that through other means, e.g. new patents for the same evolving devices, brand-name, insider-secrets, more complex/evolving interfaces/functionality that is harder to copy, etc. However, they are all trading with high valuations. 

Edited May 18, 2021 by LearningMachine

[Jurgis]

2 hours ago, LearningMachine said:

Got it.  One company I've been watching in biotech space seems to be operating more as an investment fund.  Similar to how Sequoia buys tech companies early and takes them through IPO, this company takes drug pipelines in early stages and takes them through approval, which looks like a high ROI activity to me. The company is Biogen.  It is not super-great how they have been handling their aducanumab investment, and we also don't know how well their recent other investments will actually pan-out.  Have you looked into it? 

XBI might be able to capture some of these early pipeline companies also, but then we have to study the success rate of these companies - because of the huge pricing power, getting approval is such a huge success that maybe a low success rate could be ok?

My other worry with biotech companies is that it needs just a small tweak in the law for their pricing power to deteriorate, e.g. if Medicare is allowed to negotiate with drug companies.

The other issue is constant expiry of monopoly on drugs.  Medical device companies can get around that through other means, e.g. new patents for the same evolving devices, brand-name, insider-secrets, more complex/evolving interfaces/functionality that is harder to copy, etc. However, they are all trading with high valuations. 

I am wrong person to ask about biotech. Every time I tried to get into it, I found out that the amount of material is huge and it is very hard to get a deep insight into it. I think I'm gonna defer to FSPHX for old med/pharma and XBI for the new bio.

I agree with your comments about valuations, pricing power risks, expirations, etc.