Science and Technology links (March 16th 2019)

1. There is mounting evidence that clearing old cells (senescent cells) from old tissues has a rejuvenating effect. There are very few such cells in most cases, but they cause a disproportionate amount of problems. Anderson et al. find that this effect might even extend to our hearts…
   

clearance of senescent cells in mice alleviates detrimental features of cardiac ageing, including myocardial hypertrophy and fibrosis.

Cells in our hearts do not divide very much or at all, but the authors find that they still “age” due to genetic damage.

1. Dyakonov, a famous physicist, is quite critical of the notion that quantum computers might be the future of computing:

The huge amount of scholarly literature that’s been generated about quantum-computing is notably light on experimental studies describing actual hardware. (…) There is a tremendous gap between the rudimentary but very hard experiments that have been carried out with a few qubits and the extremely developed quantum-computing theory, which relies on manipulating thousands to millions of qubits to calculate anything useful. That gap is not likely to be closed anytime soon.

We should always be skeptical regarding negative predictions. However, Dyakonov indicates that there is overwhelming emphasis on theory at the expense of practice. This is often a bad sign: theory is useful but disconnected theory is easily overrated.

• The notion that the pharmaceutical industry is driven by innovation should be regarded with skepticism. Indeed, the number of companies engaging in drug development today is lower than at any time since 1914.
• Bell et al. try to identify what turns kids into inventors:
  

(…) exposure to innovation during childhood has significant causal effects on children’s propensities to invent. Children whose families move to a high-innovation area when they are young are more likely to become inventors.

In other words, if you want your kids to be inventors, expose them to inventors and inventions.

• What happens when governments increase spending in research? Goolsbee reports that the immediate effect is to raise the wages of scientists and engineers:
  

The majority of R&D spending is actually just salary payments for R&D workers. Their labor supply, however, is quite inelastic so when the government funds R&D, a significant fraction of the increased spending goes directly into higher wages. (…) government R&D spending raises wages significantly

• Chollet comments on the current state of artificial intelligence regarding natural languages:
  

At this point we are pretty close to training our language models on the entirety of text corpora that humanity has produced so far. Their language understanding capabilities are still close to zero.

What I believe Chollet means is that by merely processing more data, and using more CPU cycles, we will not make qualitatively significant gains in artificial intelligence.

There is a counterpoint to this pessimism: it is much easier to innovate in radical ways when you have access to the right infrastructure.

• Ashworth and Ransom reports that the “college wage premium” (the financial gain derived from a college degree) is no longer increasing over time and may even be declining. Meanwhile, in the United States, the cost of attending college has tripled in a generation.
• Up until a couple of years ago, machine-learning algorithms could not play decently many of the old-school video games like Pitfall. This is no longer true: systems like Go-Explore can defeat these games. Part of the progress is due to the ability of a system like Go-Explore to receive bits of “domain knowledge” (from human beings).
• Human beings go from one relationship to another. Johnson and Neyer find that human beings tend to fall into a recurring pattern (the relationship is great adn then it deteriorate):
  

When looking at overall change from Time 1 in partnership 1 to Time 2 of partnership 2, there were no mean-level changes in relationship and sexual satisfaction, perceptions of relational instability, and frequency of conflictual and intimate exchanges (…) all constructs showed significant deterioration as the first partnership drew to a close, marked improvements as individuals moved from the end of the first partnership into their next union, and worsened across the first year of the second partnership.

• Google engineers do speech recognition using an 80MB model. This is part of a larger trend to do the computations on the small devices (e.g., smartphone) that we carry with us. To achieve such goals, we need highly efficient algorithms. The reward is lower latency (smaller wait times) and better privacy.
• Pretis finds that neither the carbon taxes nor the carbon price and emissions trading schemes introduced in Canada has reduced carbon emissions.