24 years ago I was a graduate student studying earth and planetary science at Harvard. My advisor, Professor Mike McElroy, was an expert on atmospheric chemistry and I helped him teach a course to 300 undergrads in what was known as the ‘core curriculum’. As a special treat for the students, McElroy invited his friend, Senator Al Gore, to give a talk.

Gore delivered an explosive presentation about the greenhouse effect and climate change (it amounted to an early draft of his film An Inconvenient Truth). At a reception for Gore in the department common room afterwards I got a feel for how important the subject was to my professor and his fellow academics. The 1988 Montreal Protocol to address manmade ozone depletion was a template for what was to come: the science of global warming was becoming established, and a massive global policymaking campaign would soon follow.

The gossip I overheard was that $1.1 billion was about to be appropriated by Congress for climate research over the next decade. “Get your NSF applications in now,” someone said. For meteorologists, chemists, oceanographers and other scientists dependent on government funding, the bandwagon was about to become a gravy train.

Then I attended a talk by MIT professor Richard Lindzen, a brown-bearded expert on cloud physics. He was known as the ‘prince of darkness’ in the Harvard planetary science department and he spent his lecture debunking what he argued were exaggerated global warming predictions. Lindzen seemed particularly hostile to Gore whom he described as a ‘demagogue’.

I was reminded of this episode when I read last week’s report by the International Panel on Climate Change, the scientific body set up to advise policymakers on global warming. The report was the fifth published by the IPCC and some of the characters I met at Harvard are still around. Al Gore shared a Nobel peace prize with the IPCC at the time of its fourth report in 2007. Lindzen has been adopted by climate sceptics and recently lambasted the IPCC’s latest document as vociferously as I remember him doing 24 years ago.

What’s interesting is the difference between the fourth and fifth reports (called AR4 and AR5 by the IPCC). Six years ago, AR4 trumpeted a ‘robust finding’ that global mean surface temperatures would increase in the near term by 0.2°C per decade. Last week, AR5 conceded that the actual increase per decade observed over the past 15 years was only 0.05°C, in what it delicately calls a ‘hiatus’ in the predicted trend. Expressed statistically, the observations fell outside the 95^th percentile of the range of temperature predictions ““ an embarrassing signal of model failure. (the accompanying chart taken from AR5 shows model predictions in grey and observations in red)

Although this failure of prediction has been seized upon by the sceptics, it’s only a part of the global warming evidence assembled in AR5. For example, even including the hiatus, the last 30 years were the warmest such period in eight centuries. The IPCC also identifies increasing variance in climate which is arguably more important than trends in averages, because it increases the likelihood of extreme weather events. However, the failure to get the trend in the average right seems to have chastened the IPCC, and this can be seen in the two reports’ respective treatment of uncertainty.

There are three major types of uncertainty in climate modelling. The historical state of the climate system is uncertain because of observational gaps and errors. The myriad physical, chemical and biological mechanisms that fit together in climate models are uncertain. Finally, the predictions that come out of models based on uncertain mechanisms and uncertain inputs are also uncertain.

Starting in AR4, the IPCC adopted a two-track approach to describe uncertainty. For stuff it was relatively sure about, it proposed a quantitative scale of beliefs in different findings. For example, the assertion that the last 30 years were the warmest in eight centuries is ‘very likely’ according to the IPCC which means at least a 90 percent chance of being true. Then there is a more qualitative measure of belief in a finding based on the consistency of evidence and agreement among scientists, ranging from ‘very high confidence’ to ‘very low confidence’.

By counting the number of times the different levels of confidence appear in the text of AR4 and AR5, you get a striking impression of how the belief of IPCC scientists has changed in the past six years (see histogram).

In 2007, when the IPCC and Al Gore won their Nobel prize, only 7% of qualitative assessments were ‘low confidence’ or worse. By 2013, that proportion had increased to 28%.

An example of the IPCC’s newfound caution appears in the section discussing the notorious warming hiatus. The scientists think that the hiatus was caused by a combination of two things ““ natural variability in the climate (the oceans absorbed more heat than expected) and a downturn in solar forcing of the atmosphere connected to sunspot cycles. But the scientists’ belief in this explanation is characterised as ‘low confidence’!

At least this is healthier than the cockiness that infected AR4, and the IPCC seems to make a coded admission of this when it says in AR5: “In the case of expert judgments there is a tendency towards overconfidence both at the individual level and at the group level as people converge on a view and draw confidence in its reliability from each other”.

While much of the science of global warming is sound, the IPCC scientists have learned the hard way that only by putting their doubts and disagreements on display can they fight the impression that they are lobbyists for their own gravy train.