Alcove

Science, culture, complexity

Tag: Microsoft Research

  • Who funds quantum research?

    An odd little detail in a Physics World piece on Microsoft’s claim to have made a working topological qubit:

    Regardless of the debate about the results and how they have been announced, researchers are supportive of the efforts at Microsoft to produce a topological quantum computer. “As a scientist who likes to see things tried, I’m grateful that at least one player stuck with the topological approach even when it ended up being a long, painful slog,” says [Scott] Aaronson.

    “Most governments won’t fund such work, because it’s way too risky and expensive,” adds [Winfried] Hensinger. “So it’s very nice to see that Microsoft is stepping in there.”

    In drug development, defence technologies, and life sciences research, to name a few, we’ve seen the opposite: governments fund the risky, expensive part for many years, often decades, until something viable emerges. Then the IP moves to public and private sector enterprises for commercialisation, sometimes together with government subsidies to increase public access. With pharmaceuticals in particular, the government often doesn’t recoup investments it has made in the discovery phase, which includes medical education and research. An illustrative recent example is the development of mRNA vaccines; from my piece in The Hinducriticising the medicine Nobel Prize for this work:

    Dr. Kariko and Dr. Weissman began working together on the mRNA platform at the University of Pennsylvania in the late 1990s. The University licensed its patents to mRNA RiboTherapeutics, which sublicensed them to CellScript, which sublicensed them to Moderna and BioNTech for $75 million each. Dr. Karikó joined BioNTech as senior vice-president in 2013, and the company enlisted Pfizer to develop its mRNA vaccine for COVID-19 in 2020.

    Much of the knowledge that underpins most new drugs and vaccines is unearthed at the expense of governments and public funds. This part of drug development is more risky and protracted, when scientists identify potential biomolecular targets within the body on which a drug could act in order to manage a particular disease, followed by identifying suitable chemical candidates. The cost and time estimates of this phase are $1billion-$2.5 billion and several decades, respectively.

    Companies subsequently commoditise and commercialise these entities, raking in millions in profits, typically at the expense of the same people whose taxes funded the fundamental research. There is something to be said for this model of drug and vaccine development, particularly for the innovation it fosters and the eventual competition that lowers prices, but we cannot deny the ‘double-spend’ it imposes on consumers — including governments — and the profit-seeking attitude it engenders among the companies developing and manufacturing the product.

    Quantum computing may well define the next technological revolution together with more mature AI models. Topological quantum computing in particular — if realised well enough to compete with alternative architectures based on superconducting wires and/or trapped ions — could prove especially valuable for its ability to be more powerful with fewer resources. Governments justify their continuing sizeable expense on drug development by the benefits that eventually accrue to the country’s people. By all means, quantum technologies will have similar consequences, following from a comparable trajectory of development where certain lines of inquiry are not precluded because they could be loss-making or amount to false starts. And they will impinge on everything from one’s fundamental rights to national security.

    But Hensinger’s opinion indicates the responsibility of developing this technology has been left to the private sector. I wonder if there are confounding factors here. For example, is Microsoft’s pursuit of a topological qubit the exception to the rule — i.e. one of a few enterprises that are funded by a private organisation in a sea of publicly funded research? Another possibility is that we’re hearing about Microsoft’s success because it has a loud voice, with the added possibility that its announcement was premature (context here). It’s also possible Microsoft’s effort included grants from NSF, DARPA or the like.

    All this said, let’s assume for a moment that what Hensinger said was true of quantum computing research in general: the lack of state-led development in such potentially transformative technologies raises two (closely related) concerns. The first is scientific progress, especially that it will happen behind closed doors. In a June 2023 note, senior editors of the Physical Review B journal acknowledged the contest between the importance of researchers sharing their data for scrutiny, replication, and for others to build on their work — all crucial for science — and private sector enterprises’ need to protect IP and thus withhold data. “This will not be the last time the American Physical Society confronts a tension between transparency and the transmission of new results,” they added. Unlike in drug development, life sciences, etc., even the moral argument that publicly funded research must be in the public domain is rendered impotent, although it can still be recast as the weaker “research that affects the public sphere…”.

    The second is democracy. In a March 2024 commentary, digital governance experts Nathan Sanders, Bruce Schneier, and Norman Eisen wrote that the state could develop a “public AI” to counter the already apparent effects of “private AI” on democratic institutions. According to them, a “public AI” model could “provide a mechanism for public input and oversight on the critical ethical questions facing AI development,” including “how to incorporate copyrighted works in model training” and “how to license access for sensitive applications ranging from policing to medical use”. They added: “Federally funded foundation AI models would be provided as a public service, similar to a health care private option. They would not eliminate opportunities for private foundation models, but they would offer a baseline of price, quality, and ethical development practices that corporate players would have to match or exceed to compete.”

    Of course, quantum computing isn’t beset by the same black-box problem that surrounds AI models, yet what it implies for our ability to secure digital data means it could still benefit from state-led development. Specifically: (i) a government-funded technology standard could specify the baseline for the private sector to “match or exceed to compete” so that computers deployed to secure public data maintain a minimum level of security; (ii) private innovation can build on the standard, with the advantage of not having to lay new foundations of their own; and (iii) the data and the schematics pertaining to the standard should be in the public domain, thus restricting private-sector IP to specific innovations.[1]

    [1] Contrary to a lamentable public perception, just knowing how a digital technology works doesn’t mean it can be hacked.

  • Scientists’ conduct affects science

    Nature News has published an excellent feature by Edwin Cartlidge on the “wall of scepticism” that arose in response to the latest superconductivity claim from Ranga Dias et al., purportedly in a compound called nitrogen-doped lutetium hydride. It seems the new paper has earned a note of concern as well, after various independent research groups failed to replicate the results. Dias & co. had had another paper, claiming superconductivity in a different material, retracted in October 2022, two years after its publication. All these facts together raise a few implications about the popular imagination of science.

    First, the new paper was published by Nature, a peer-reviewed journal. And Jorge Hirsch of the University of California, San Diego, told Nature News “that editors should have first resolved the question about the provenance of the raw data in the retracted 2020 Nature article before even considering the 2023 paper”. So the note reaffirms the role of peer-review being limited to checking whether the information presented in a paper is consistent with the paper’s conclusions, and not checking whether it is well-founded and has integrity in and of itself.

    Second, from Nature News:

    “Researchers from four other groups, meanwhile, told Nature’s news team that they had abandoned their own attempts to replicate the work or hadn’t even tried. Eremets said that he wasted time on the CSH work, so didn’t bother with LuNH. ‘I just ignored it,’ he says.”

    An amusing illustration, I think, that speaks against science’s claims to being impartial, etc. In a perfectly objective world, Dias et al.’s previous work shouldn’t have mattered to other scientists, who should have endeavoured to verify the claims in the new paper anew, given that it’s a fairly sensational claim and because it was published in a ‘prestigious’ peer-reviewed journal. But, as Eremets said, “the synthesis protocol wasn’t clear in the paper and Dias didn’t help to clarify it”.

    The reciprocal is also true: Dias chose to share samples of nitrogen-doped lutetium hydride that his team had prepared only with Russell Hemley, who studies material chemistry at the University of Illinois, Chicago, (and some other groups that he refused to name) – and that Hemley is one of the researchers who hasn’t criticised Dias’s findings. Hemley is also not an independent researcher; he and Dias worked together on the work in the 2020 paper that was later retracted. Dias should ideally have shared the samples with everyone. But scientists’ social conduct does matter, influencing decisions about how other scientists believe they should respond.

    Speaking of which: Nature (the journal) on the other hand doesn’t look at past work and attendant misgivings when judging each paper. From Nature News (emphasis added):

    The editor’s note added to the 2023 paper on 1 September, saying that the reliability of data are in question, adds that “appropriate editorial action will be taken once this matter is resolved.” Karl Ziemelis, Nature’s chief applied- and physical-sciences editor, based in London, says that he and his colleagues are “assessing concerns” about the paper, and adds: “Owing to the confidentiality of the peer-review process we cannot discuss specific details of what transpired.” As for the 2020 paper, Ziemelis explains that they decided not to look into the origin of the data once they had established problems with the data processing and then retracted the research. “Our broader investigation of that work ceased at that point,” he says. Ziemelis adds that “all submitted manuscripts are considered independently on the basis of the quality and timeliness of their science”.

    The refusal to share samples echoes an unusual decision by the journal Physical Review B to publish a paper authored by researchers at Microsoft, in which they reported discovery a Majorana zero mode – an elusive particle (in a manner of speaking) that could lead the way to building quantum ‘supercomputers’. However, it seems the team withheld some information that independent researchers could have used to validate the findings, presumably because it’s intellectual property. Rice University physics professor Douglas Natelson wrote on his blog:

    The rationale is that the community is better served by getting this result into the peer-reviewed literature now even if all of the details aren’t going to be made available publicly until the end of 2024. I don’t get why the researchers didn’t just wait to publish, if they are so worried about those details being available.

    Take all of these facts and opinions together and ask yourself: what then is the scientific literature? It probably contains many papers that have cleared peer-review but whose results won’t replicate. Some papers may or may not replicate but we’ll never know for a couple years. It also doesn’t contain replication studies that might have been there if the replicators and the original research group were on amicable terms. What also do these facts and view imply for the popular conception of science?

    Every day, I encounter two broad kinds of critical imaginations of science. One has emerged from the practitioners of science, and those studying its philosophy, history, sociology, etc. These individuals have debated the notions presented above to varying degrees. But there is also a class of people in India that wields science as an antidote to what it claims is the state’s collusion with pseudoscience, and such collusion as displacing what is apparently science’s rightful place in the Indian society-state: as the best and sole arbiter of facts and knowledge. This science is apparently a unified whole, objective, self-correcting, evidence-based, and anti-faith. I imagine this science needs to have these characteristics in order to effectively challenge, in the courts of public opinion, the government’s oft-mistaken claims.

    At the same time, the ongoing Dias et al. saga reminds us that any ‘science’ imprisoned by these assumptions would dismiss the events and forces that would actually help it grow – such as incentivising good-faith actions, acknowledging the labour required to keep science honest and reflexive, discussing issues resulting from the cultural preferences of its exponents, paying attention to social relationships, heeding concerns about the effects of one’s work and conduct on the field, etc. In the words of Paul Feyerabend (Against Method, third ed., 1993): “Science is neither a single tradition, nor the best tradition there is, except for people who have become accustomed to its presence, its benefits and its disadvantages.”