Special software with machine-learning algorithms will identify potential wrongdoings in public procurement.
On Thursday, 1 November, Transparency International Ukraine presented a beta version of the new system of civic control over public procurement DOZORRO.
It is based on special software which learns to identify tenders with a high risk of corruption. According to the IT portal developer Volodymyr Flionts, what motivated him to create the software was the question that ProZorro creators are often asked: why doesn’t the system punish the corrupt? So the team started working on a software solution that would help to find and identify them.
DOZORRO artificial intelligence is vastly different from risk indicators. of the State Audit Service of Ukraine. On 30 October 2018, the State Audit Service started automatic verification of tenders in ProZorro based on 35 risk indicators. They include violations of the bidding procedure, bidding participants not having powers to participate, absence of digital signatures, violations of review deadlines, absence of the concluded agreement, absence of uploaded tender documentation, absence of the description of the procured item, a large quantity of lots included in one tender etc. The more indicators are flagged, the greater is the chance that the tender will be verified by the auditors. The DOZORRO system, on the other hand, is more flexible and does not have an exhaustive list of indicators.
“Once there are clear criteria of a potentially suspicious tender, corrupt officials easily adapt to them, since they are especially interested in finding loopholes to avoid crossing any lines. That is why our system does not have any definite coefficients and formulas which procuring entities could use to dishonest ends. As soon as experts and activists see that new “trickery algorithms” have emerged, the system will automatically adjust,” explains Volodymyr Flionts.
Artificial intelligence started learning to identify violations in procurement in July of 2018. The developers sent 20 experts about 3,500 tenders to check. They were asked to answer one single question – whether they have any risks or not. Experts did not know the amounts or the names of the procuring entities to make their assessment as objective as possible. All responses were then fed into the AI algorithm.
Now the system assesses the likelihood of corruption risks in tenders independently and then sends them to civil society organizations of the DOZORRO community. If the suspicions were correct, the software remembers its choice, if they were wrong, forgets it. This way, the AI algorithm’s assessment of potentially risky tenders becomes increasingly accurate.
90% Accuracy
DOZORRO product manager Andrii Kucherenko shared the beta test results: thanks to the algorithm, 26% more tenders with unfounded selection of the winner were identified, 37% more tenders with groundless disqualification, 298% more with participants’ conspiracy. It is noteworthy that the AI found the most violations in the most expensive tenders.
The developers say that no software will ever be able to identify violations in tenders better than experts do. But AI algorithms will make their work much more efficient. The program is expected to be fully put into operation in the first quarter of 2019. The DOZORRO artificial intelligence will enable activists to identify trickery in public procurement much faster and report their findings to supervisory and law enforcement agencies.
