The ideal group to review this topic would be Senior Software Architects and Technology Strategists with a background in foundational computer science, particularly functional programming and historical systems analysis.
Abstract:
This keynote addresses the inherent difficulty of technological prediction, specifically challenging the predictive validity of models like the Gartner Hype Cycle. Drawing upon a career beginning with Lisp in 1971 and including work on the Maxima computer algebra system, the speaker provides historical context for both celebrated and failed technological predictions. Key arguments focus on the unexpected pathways to technological success—often through peripheral functionality or indirect influence (Greenbaum’s Tenth Rule)—rather than the core design purpose. Case studies include the Maxima system, where symbolic solutions proved less practical than numerical methods, the failure of the Architecture Neutral Distribution Format (ADMF) due to market dynamics, and the premature prediction of the internet’s mass adoption. The concluding advice emphasizes embracing uncertainty, developing minimum viable products, and recognizing that timing is often the most unpredictable factor in technological maturation.
Is the Hype Cycle Real?
0:14 Introduction and Heritage: The speaker, Stavros Macrakis, identifies his long history with Lisp, beginning in 1971. He notes his association with the Maxima computer algebra group at MIT, which is presented as the primary reason for his invitation.
0:45 Keynote Topic: Prediction Difficulty: Introduces the central theme that predictions, especially about the future, are difficult, supported by research (Raifa, 1969) indicating people are dramatically overconfident in their confidence intervals (50% error rate when expecting 10%).
10:21 The Gartner Hype Cycle Critique: The speaker describes the Gartner Hype Cycle (Technology Trigger, Peak of Inflated Expectations, Trough of Disillusionment, Plateau of Productivity) but notes a study by Melany found that only 50 out of 200 listed technologies reappeared the following year, challenging the cycle's predictive power.
12:55 Notable Failed Predictions: Enumerates several technologies that failed to meet expectations, including Desktop Linux, RSS in the Enterprise, Mesh Networking (for home use), and Japan’s Fifth Generation Computing project (based on Prolog).
15:48 Failure of Micro Payments: Details the failure of micro payment systems, arguing the issue was psychological, not technical. Consumers prefer predictable subscriptions (like newspapers and mobile phone plans) over per-use charges.
18:01 The Gold Rush Analogy Applied: References the California Gold Rush analogy (sell shovels, not mine gold), cautioning that developing core tools (like compilers) does not typically lead to financial success.
19:34 Lisp Success Defined: Questions the definition of Lisp's success, noting it hasn't generated significant profit but highlights Greenbaum's Tenth Rule—that complex C/Fortran programs often contain a flawed implementation of half of Common Lisp, indicating indirect adoption.
24:28 Early MIT AI Lab as a Predictive Model: Recounts the advanced state of computing at the MIT AI Lab in the 1960s/70s (email, distribution lists, chat, structured documents like Scribe, 1980), noting that many internet components were available years before the widespread adoption of the World Wide Web in the 1990s.
27:53 Surprises in Complexity Solving: Identifies NP-complete problems (specifically 3-SAT) as an area where prediction failed positively; despite theoretical complexity, modern software routinely solves problems with tens or hundreds of thousands of clauses in practice.
31:49 Maxima Symbolic Math Limitations: Discusses the failure of symbolic math systems like Maxima to be practically useful for complex problems (e.g., quartic equation solving). Symbolic results become too complicated to interpret and are often numerically unstable compared to faster, more accurate numerical solvers.
36:37 Maxima's Unexpected Utility: Notes that Maxima’s concept of a matrix as an array of formulas allowed one user to invent a spreadsheet functionality before commercial spreadsheets existed, illustrating that peripheral or non-core capabilities often prove most valuable.
40:02 The ADMF Product Management Failure: Details his experience as a Product Manager at the Open Software Foundation (OSF) overseeing the Architecture Neutral Distribution Format (ADMF). The technology, designed to allow software to run across competing architectures (IBM, HP, Digital), was technically sound but failed because market leaders (Digital) opposed the standardization that would help smaller competitors, and ISVs found the primary cost was QA, not porting.
45:34 Market Dynamics Over Technical Superiority: Concludes the ADMF story by noting that the technology was rendered obsolete by the market consolidation around x86 architecture, an event (Microsoft/commodity hardware) that rendered the need for ADMF irrelevant.
46:11 Logical Deduction and Legal Systems: Discusses the application of logical systems (like Prolog) to complex laws (e.g., British nationality code). Lawyers found the system useless because their practice focuses on interpreting ambiguous terms in the law, not just applying clear logic.
50:04 Conclusion on Prediction and Strategy: Reaffirms that prediction is hard, and success often comes from a sequence of technology derivatives. The critical missing factor in predictions is timing. Recommended strategy: embrace uncertainty, work on what is interesting/fun, pivot when necessary, and develop the Minimum Viable Product (MVP) to gauge unpredictable user response.