Recent weeks have brought good news about the criminal-justice system in this country. Despite hard times and dire predictions to the contrary, crime continues to decline. And the Pew Center for the States reports that the prison population has actually shrunk for the first time in more than a generation.

As is often the case, however, the closer one looks, the more complicated the picture becomes. Crime may be down in lots of places, but there are many urban, low-income neighborhoods that remain fundamentally unsafe. And while it's no longer on the rise, the U.S. incarceration rate still well exceeds that of most industrialized nations.

How are we to fix such deeply rooted problems? Simply maintaining the status quo isn't good enough. We must aggressively seek out new ideas.

Unfortunately, though, there is a major obstacle to innovation in the criminal-justice system: fear of failure. Many prosecutors, police chiefs, and court administrators don't feel they have the freedom to engage in trial and error. Unlike some other fields, criminal justice does not have a rich tradition of honest discussion of mistakes.

The pharmaceutical company Eli Lilly used to hold "failure parties" to acknowledge outstanding scientific work even when it failed to lead to successful new products. Needless to say, there are no failure parties in criminal justice.

What would we learn if criminal-justice officials looked at their work with the probing, skeptical eye of a scientist? First, we would see that there are many different reasons why new programs fail.

Sometimes the problem is conceptual: Some ideas just aren't good. For example, research tells us that gun buyback programs, in which financial incentives are offered in exchange for surrendered guns, have done very little to stem gun violence.

Similarly, programs such as DARE, which send police officers into schools to counsel teens on the dangers of drugs, have had no discernible impact on substance abuse. Nor is there any evidence that "scared straight" programs, which give young people tours of prisons, reduce delinquency.

But bad ideas aren't usually the problem. More often, initiatives fail because of the details of their implementation. The criminal-justice system is an enormously complicated machine, and if reformers make one false move, it has a tendency to revert to form.

Drug courts are a case in point. In contrast to the vast majority of criminal-justice programs, there is solid evidence that drug courts - which link addicted offenders to monitored treatment instead of prison - actually make a difference, reducing both substance abuse and recidivism.

But this doesn't mean that every drug court is a success. Some have struggled mightily because of a variety of implementation errors, including failing to effectively market the program to decision-makers and to plan for transitions between judges.

Drug courts are just one example of how difficult it is to make categorical judgments about the success of a new initiative. Another comes from Harlem, where the Center for Court Innovation helped conceive and implement a reentry program providing enhanced services and supervision during the first six months of parole.

We were discouraged when a recent evaluation found that the increased supervision by a team of parole officers, social workers, and an administrative law judge actually had more participants going back to prison for technical parole violations - missing an appointment, breaking curfew, flunking a drug test. On the other hand, the program reduced crime, with participants 19 percent less likely to be re-convicted.

A scientist would see these results not as contradictory and confusing, but as a necessary step in the process. As David Shenk notes in The Genius in All of Us, the latest research on the brain suggests it takes thousands of hours of deliberate practice - and frequent failures - to achieve greatness at any task. Why should criminal-justice reform be different?

Unless criminal-justice officials are encouraged to risk failure and given the time to engage in a learning process, we will never make a dent in our most persistent public-safety problems.