Teasing future performance from historical data
The Intelligence Advanced Research Projects Activity suggests that historical performance metrics may help make predictions about a technology's future performance in an objective and quantitative manner.
To better predict the future performance of various technologies, the Intelligence Advanced Research Projects Activity is looking for datasets documenting a technology's performance over several decades, the way the TOP500 database tracks metrics like processing speed, cores, power consumption to track the performance of supercomputers.
In a recent request for information, IARPA says it wants to analyze key technology spaces -- including cybersecurity, artificial intelligence, quantum sciences and "human performance modification" -- by analyzing developments of their constituent systems and components over time. IARPA suggests this information can be used to make predictions about future performance in an objective and quantitative manner.
As an example, IARPA cites capacitor technology as a performance metric for power density. "As capacitor technology improves, increasing power density (kilowatts per kilogram) is a good indicator of performance improvement," according to the RFI. "Ultimately historical improvements in power density may be used to predict future improvements in power density."
To test this idea, IARPA wants suggestions of constituent technologies, datasets and metrics to document the historical performance of broad technologies. Submissions should describe why the performance of a particular constituent technology, system or component can serve as a suitable measure of the broader technology.
Datasets should cover at least 10 years and include accurate metrics as well as information on when the technology was first developed. Data sources can be public or private, though non-proprietary, machine-readable and unclassified datasets are preferable. They should also be documented by a credible source or corroborated by an independent third party. Detailed references are encouraged where possible.
Suggested metrics must be objective, quantitative and available in the historical record.
Responses are due Aug. 6. Read the RFI here.