FREE NEWSLETTER: Econintersect sends a nightly newsletter highlighting news events of the day, and providing a summary of new articles posted on the website. Econintersect will not sell or pass your email address to others per our privacy policy. You can cancel this subscription at any time by selecting the unsubscribing link in the footer of each email.

posted on 03 June 2017

HLMI - Party Time or Human Extinction Time?

Random Thoughts from the High Desert

Written by

HLMI is not an Internet Protocol but stands for High-Level Machine Intelligence. In the past generally speaking, the more intelligent our domesticated animals, the better; the more intelligent the machines we created, the better. Does that remain the case? That is the topic for discussion with a question. Will robots find our National Parks an enjoyable place to visit ... even the sandy ones?

Empty Scene Kova National Park

Please share this article - Go to very top of page, right hand side for social media buttons.

When Will AI Exceed Human Performance? Evidence from AI Experts is a pretty comprehensive article about the progress of Artificial Intelligence to Exceed Human Performance.

For the methodology you can read the full paper. Here I have extracted the Abstract and some of the more interesting tables and graphics.

When Will AI Exceed Human Performance?

Evidence from AI Experts

Katja Grace 1,2 John Salvatier 2 Allan Dafoe 1,3 Baobao Zhang 3 and Owain Evans

1 Future of Humanity Institute, Oxford University

2 AI Impacts

3 Department of Political Science, Yale University


Advances in artificial intelligence (AI) will transform modern life by reshaping transportation, health, science, finance, and the military. To adapt public policy, we need to better anticipate these advances. Here we report the results from a large survey of machine learning researchers on their beliefs about progress in AI. Researchers predict AI will outperform humans in many activities in the next ten years, such as translating languages (by 2024), writing high-school essays (by 2026), driving a truck (by 2027), working in retail (by 2031), writing a bestselling book (by 2049), and working as a surgeon (by 2053). Researchers believe there is a 50% chance of AI outperforming humans in all tasks in 45 years and of automating all human jobs in 120 years, with Asian respondents expecting these dates much sooner than North Americans. These results will inform discussion amongst researchers and policymakers about anticipating and managing trends in AI.

The first two graphics below relate to when AI will reach and exceed Human Capability.

First the Overall Results

When will Machines be as Smart as People.

Breaking out the results by Nation of Participation in the Survey (where first academic degree was earned, not where currently residing).

Breakdown by the Nationality of the Survey Participant

Notice that North America is the least "optimistic" on the speed that machines will outperform humans. Is that better insight or denial?

Next the issue of when employment is eliminated

Replacement of Human Labor

Again there is a big difference between the views of North Americans versus Asians: About 65 years difference in the median time from the survey.

Assessment of the Survey

In many ways this is a very good study. Some pertinent comments:

  • All survey participants have presented on this topic at known conferences
  • 21% of the 1634 authors responded
  • A sampling of the non-responders suggest they have similar characteristics to the responders
  • To keep the survey form short, only a subset of questions were answered by each responder. This could be an issue re the second major topic namely the employment implications
  • The survey addressed 32 milestones re achievement of HLMI. Unless I missed it, it looks like the largest number of respondents re these milestones was 52 for one milestone. So the survey sample for each milestone is much smaller than the overall survey sample and the question on timing of automation of human jobs has a sample size that I was not able to ascertain from reading the report.
  • 82% of responders work in academia, 21% work in industry. In general it seems the survey was directed at those who understood HLMI but not economics. Thus the first two graphics are probably more reliable than the third i.e. the rate of improvement in machine intelligence is probably well understood by those who participated in this survey but how this translates to employment is probably less well understood and the implications perhaps even less well understood. A related oddity in the study is that the median time for HLMI-entities to outperform humans on all tasks was 45 years but the median time for HLMI-entities to outperform humans as an AI Researcher was about 85 years. The logical conclusion from that is that AI Research is not currently being performed by humans. Obviously this has to do with how the sample was segmented re the 32 Milestones and other questions.

Limitations of the Study

I have one meal a day at a Subway Restaurant. It is easy to see how a machine can make a sandwich to a customer's specifications. It is harder to see how a machine can handle a street person in the restaurant annoying customers. Even changing the paper tape in the cash register can be very complex especially if the cash register is a bit off in following the prescribed method of operation. The Subway employees are able to sort out many problems. A machine can do predefined functions and may, at a higher cost, have the ability to perform tasks not predefined. But with humans, you often get abilities that exceed the minimum applicant selection criteria for free. Thus the HLMI level needed for many jobs to deal with low-probability events may need to exceed that which you often get with the prevailing wage and the operator of the facility (private or public) may need to pay for capabilities in excess of that which comes for free with the prevailing wage. This is especially true when retired persons or the disabled take jobs at lower skill and compensation levels than they once commanded. This may not be properly captured in a survey of academics who are not totally familiar with the full spectrum of tasks that workers perform.

We talk about AI but I wonder if this is really much more than what used to be called expert systems. To what extent is progress in HLMI based on breakthroughs or simply the Learning Curve?

A hypothetical example of a complex job is a judge committing intrinsic fraud and getting away with it. Can a machine do this? It involves being able to do a stakeholder analysis? Who cares and are they able to create a problem for that judge. It seems easier to have a machine understand everything than to have a machine selectively misunderstand certain things. That inability to decide when to ignore information may impede the introduction of HLMI-entities into the economy.

I assume that we can create any desired level of HLMI. The question is the comparative cost vs. human labor. As machines do more, the need for labor will go down and the needs of this reduced number of workers for that which is not provided at reduced if not zero cost by machines goes down - so the supply curve of labor may shift to the right (also known as downward) creating generally a less expensive supply curve. This provides for stiffer competition for the HLMI-entities.

Implications of the Study

The study results are that the risk of a really bad outcome from HLMI such as human extinction is 5%. I think the risk is closer to 100% for Planet Earth. If people are no longer needed it seems to me that intelligent machines will find humans annoying and eliminate them. Continuing with my unsubstantiated opinion, I doubt they would track down humans throughout the universe unless those living somewhere other than Planet Earth were perceived as a threat to the HLMI-entities on Planet Earth.

A society where no one needs to work but all are dependent on machines is a society which:

  • May be boring and
  • May lend itself to domination by human elites who control the HLMI-entities or the HLMI-entities themselves.

The transition from a society where labor earns goods and services to one where some need to work but an increasing percentage of what is normally considered the labor force does not need to work probably creates serious class envy problems as the worker class shrinks in size and may not be able to be recruited from volunteers. The jobs not able to be performed by HLMI-entities may not be fun and may be very stressful. Coercion may be needed to fill those jobs.

Examples of possibly unanticipated implications would include if machines could write a seemingly valid high-school essay ten years from now, essay assignments would seem to no longer be able to play a role in high school education. If truck drivers start to be replaced by HLMI-entities in 12 years, there will be a big drop off in demand for motel rooms and other services that are based on meeting the needs of human truckers. This would result in fewer highway services for other travelers. Many small highway-oriented towns might die.


This is very good study. The timing reported in the study may be on target or not but the process appears to be inexorable. When I worked in mining there was a rule that one could not have a cartel if there were as many as 11 producers. At about the point, cheating is perceived as being more advantageous than cooperation. OPEC is a good example of this. Thus I believe it is unlikely that humans will cooperate to prevent HLMI-entities from getting too smart and being able to have their own opinion about things. Thus it seems to me that HLMI is far more significant than Global Warming which if allowed to go unchecked would begin to have serious implications three centuries from now. HLMI might eliminate the human race in two centuries or even one.

Regarding the question ofwhether HLMI-entities find National Parks of interest, I leave that as an open question with unknown implications.


>>>>> Scroll down to view and make comments <<<<<<

Click here for Historical News Post Listing

Make a Comment

Econintersect wants your comments, data and opinion on the articles posted. You can also comment using Facebook directly using he comment block below.

Econintersect Contributors

Print this page or create a PDF file of this page
Print Friendly and PDF

The growing use of ad blocking software is creating a shortfall in covering our fixed expenses. Please consider a donation to Econintersect to allow continuing output of quality and balanced financial and economic news and analysis.

Keep up with economic news using our dynamic economic newspapers with the largest international coverage on the internet
Asia / Pacific
Middle East / Africa
USA Government



Analysis Blog
News Blog
Investing Blog
Opinion Blog
Precious Metals Blog
Markets Blog
Video of the Day


Asia / Pacific
Middle East / Africa
USA Government

RSS Feeds / Social Media

Combined Econintersect Feed

Free Newsletter

Marketplace - Books & More

Economic Forecast

Content Contribution



  Top Economics Site Contributor TalkMarkets Contributor Finance Blogs Free PageRank Checker Active Search Results Google+

This Web Page by Steven Hansen ---- Copyright 2010 - 2017 Econintersect LLC - all rights reserved