EU missed artificial intelligence targets due to Horizon Europe delay, auditors find

The European Court of Auditors (ECA) has issued a stinging report criticising the European Commission’s AI strategy, finding that the EU underspent on AI by €600 million because of a delay in 2021 starting Horizon Europe.

The ECA took aim at a range of failings, including a lack of checks on projects after they had ended, the fragmented alphabet soup of departments and agencies responsible for AI policy, and a lack of an “accurate overview” of the AI projects funded by the EU.

“It's a matter of concern […] that more than five years after the adoption of the EU AI plan, weaknesses in implementation and performance monitoring show the need for the Commission's increased focus on delivering results,” said Mihails Kozlovs, the ECA member who led the review, in a press conference yesterday.

The Commission, however, has defended its record and rejected an unusually high number of the proposals put forward by the ECA in its report.

The ECA’s intervention comes at a crucial time, as the EU still shows no sign of catching up with the US or China in a technology seen as key to economic growth and security. The bloc still has a strong position in AI research, but lags far behind on metrics of patents and private investment.

Last year, the US reported €55 billion in venture capital funding for AI, China €18 billion, and the EU just €8 billion, a sizable drop on 2022, according to figures from the Organisation for Economic Co-operation and Development.

The audit stretches back to 2018, when European governments started releasing a flurry of AI plans to capitalise on the technology’s potential to drive growth. In that year, the Commission created its own AI plan, which it updated in 2021.

Part of the strategy was to pump money into AI R&D through the Horizon 2020 and Horizon Europe research and innovation programmes.

The Commission nearly hit its spending targets of €1.5 billion in the 2018-2020 period, but fell €600 million short of a €2 billion target for 2021-22, “due to delays in adopting the Horizon Europe programme, which is the main source of AI project financing in the EU”, the report found.

The auditor’s report was an “accurate assessment” that “in the fast-paced field of AI, the EU is doing too little too late on AI innovation,” said Robert Praas, an AI specialist at the CEPS think tank in Brussels.

“Political will is not translated into implementation,” he said. “My guess is that the ambition is not paired with enough staff capable of executing it: being a regulator is different from an innovator”.

No coordination

The auditors also warned that there was “no EU body or committee to coordinate the projects at the planning, implementation or evaluation stages.”

This led to overlapping projects, it said. For example, the EU funded three different AI taxonomy projects, one through Horizon 2020, another through the European Institute of Innovation & Technology (EIT), and another with the Joint Research Council, “without there being any coordination between them”.

There is a “high level of fragmentation of AI funding and management,” the report says. The auditors reel off a voluminous list of EU bodies responsible, from Commission departments including its research and information technology directorates-general,  to multi-partner joint undertakings, executive agencies, plus the EIT.

While Horizon projects are tracked during their lifetime, the auditors criticised the Commission’s failure to check afterwards whether the work had been commercially exploited.

In some cases, research never left the lab, despite incredibly detailed plans for real world exploitation, the auditors found.

Failed exploitation plans

One cybersecurity project, focusing on autonomous cars, had an exploitation plan which ballooned from eight pages in the project proposal to 117 pages in its final plan.

“However, the project did not result in any commercialisation of results, and the Commission did not have evidence of any continuation of the project at the time of the audit,” the ECA found.

Such complex plans do “not necessarily lead to actual commercialisation or exploitation results,” the auditors said. By contrast, the US National Science Foundation asks grantees for only a short dissemination plan with their proposal, they pointed out.

Per €10 million spent on AI in Horizon 2020, inventors applied for an average of 0.65 patents, the audit found, far below targets, and below the programme average of 0.87.

There was also concern that the Commission did not have an “accurate overview” of the AI projects it was funding, as there was no “systematic tagging” of projects until 2020. That however has changed since Horizon Europe started in 2021.

CEPS’s Praas agreed that searching for EU AI projects was too difficult, and that certain databases were not up to date.

Commission response

In its response to the ECA, the Commission defended its record on AI, and pointed to the establishment of a new European AI Office – which officially opens today -  which will help enforce the bloc’s new AI Act.

The ECA’s audit only officially covers the period up to 2022, and since then, the Commission said more support schemes had gotten up and running. The European Innovation Council has pumped €440 million in grants and investments to companies “deploying or developing AI” in the period 2021-23, it said.

On the ECA’s specific recommendations, the Commission agreed to several, including to update its targets for AI investment in light of the boom in generative AI. The auditors complained that the Commission had not updated them since they were set in 2018.

But the Commission rejected the ECA’s demand that it should create a framework for better tagging and tracking financial support for AI in the EU, arguing that it was too expensive and burdensome.

Tracking support for existing prioritise like climate, biodiversity and gender equality was already “a work-intensive process”, it argued, and tracking new categories would “increase information requirements and place a greater burden on beneficiaries, Member States, and implementing entities”.

It also rejected the ECA’s calls to set out AI-specific performance targets and indicators, pointing out that Horizon Europe and Digital Europe, which provide the bulk of support for AI projects, already have indicators and reporting requirements attached.

“Introducing AI-specific indicators might not be relevant in view of the fast-evolving nature of the related technologies and innovation,” it said.