Cheap Computation Abounds, but Coordination Lags

Raw computation, once a primary business constraint, has seen its price decrease faster than most other economic resources, noted Dr. Aditya Vikram Kashyap, an AI and Innovation Leader. This has made problem-solving cheaper. However, organisations are not becoming effective at a pace equal to their tools' reduced cost. The gap between what a company can calculate and accomplish now widens, indicating a new, underlying constraint.

Credit: Cloudinary

OpenAI, for example, introduced its first commercial language model to developers, initially charging US$60 to process a million tokens. Stanford's Human-Centered AI Institute subsequently reported a substantial cost reduction. The price for querying a model at GPT-3.5's performance level on the MMLU benchmark dropped from US$20 per million tokens to US$0.07 over approximately 18 months. This more than 280-fold decrease shifted advanced analytics from a research budget necessity to a subscription-based offering.

But, despite this abundance of cheap computational power, many organisations struggle to translate it into improved overall performance. The issue stems from confusing two distinct problems: computation and coordination. Computation solves individual tasks, such as forecasting demand, pricing risk, or detecting a tumour. Coordination aligns numerous individual solutions, generated by disparate people, systems, and institutions, towards a unified outcome.

Coordination failures are particularly evident where computation is plentiful. American healthcare serves as a prime example, possessing more data and analytical capacity per US dollar of output than many other sectors. A review published in JAMA in a recent year estimated the US healthcare system wastes between US$760 billion and US$935 billion annually, a quarter of all health expenditure. Between US$27 billion and US$78 billion of this waste was attributed to inadequate care coordination, encompassing avoidable admissions, duplicated tests, and complications from aligning disparate medical actors across fragmented workflows. The diagnosis is rarely the bottleneck; the handoff is.

Supply chains echo this pattern. Manufacturers possess forecasting capabilities far surpassing previous generations. Yet, they remain vulnerable to disruption. Knowing future demand is simpler than synchronising suppliers, carriers, customs authorities, and ports across continents and varied legal frameworks. Financial markets also depend on coordinated architecture, including exchanges, clearinghouses, custodians, payment systems, and regulators, operating under shared rules to maintain trust during stress.

And, in the year 2020, the US Treasury market, one of the world's deepest and most analysed, experienced significant disruption. This breakdown did not result from a lack of computation but from a rush to sell that overwhelmed dealers' capacity to intermediate trades. Only extensive Federal Reserve intervention restored stability. Markets rarely fail for want of processing power; they fail when coordination between institutions gives way.

This highlights a fundamental aspect of modern organisational life. Institutions employ two forms of intelligence. Computational intelligence processes information to produce answers. Coordination intelligence aligns independent actors, each with their own incentives, information, and authority, towards coherent collective action. These two developed together for much of the last century but now advance at vastly different rates. Computational intelligence is a commodity. Coordination intelligence is an accumulated capability through governance, incentive structures, decision rights, interoperation standards, and trust.

So, the strategic implications are clear. The advantage derived from faster analysis is diminishing, as all participants access the same increasingly inexpensive computational resources. The lasting advantage now lies in architectural design. This belongs to organisations that meticulously craft institutional frameworks governing decision flow. These frameworks include incentives guiding independent actors, standards enabling separate systems to interoperate, and decision rights that permit choices without escalating through multiple hierarchical layers. This design requires incremental construction, not download.

Finally, a broader economic pattern supports this perspective. Each major economic development era has been defined by overcoming a single dominant constraint, from production costs in the industrial age to computation costs in the information age. The current situation mirrors this, with computation no longer the primary bottleneck for many organisations. The defining organisations of this century will master institutional design: building governance, incentives, and standards for independent actors to function as one.

• Raw computational power has become significantly cheaper, but organisational effectiveness has not risen proportionally.

• Coordination, not computation, is the primary constraint limiting a company's ability to achieve objectives.

• Coordination failures are observed across various sectors, including healthcare, supply chains, and financial markets.

• The enduring advantage for organisations will stem from architectural design and institutional coordination rather than raw processing capability.

Source: Forbes