Well, you can ask for their consensus if possible and input their agreed score.

Or, if consensus is inappropriate, you may use the median score of the respondent. However, this may require slight adjustments when considering their consistency across criteria and alternatives.

Alternatively, you may want to use this excel file. To be honest, I forgot how I came to this spreadsheet (attached) by Goepel, K. (2012).

Hope this helps.

Based on my experience, The easiest way is to calculate an average. You can see the procedure in the link below: https://www.researchgate.net/publication/233899063_Impact_of_child_labor_on_academic_performance_Evidence_from_the_program_Edcame_Primero_Colombia

Article Impact of child labor on academic performance: Evidence from...

Thank you Prof. Oscar and Prof.Imam, for your valuable inputs.

Dear Oscar,

I would like to argue that using the median is better and not that much harder than calculating average. I prefer using median since the result will most likely be an integer.

Of course this preference to integer is purely my personal opinion.

How about computing the priority matrix for each survey response and then take geometric average to aggregate the results?

Yes, using geometric mean will yield aggregates with better consistency than using median or arithmetic mean.

I think the fundamental question should be raised:

is group's opinion better than the one belonging to a single decision maker ?

For sure, the answer is very debatable.

The reason for which consensus was introduced in the foundations of AHP was to deliver a method for decreasing consistency ratio toward an acceptable value. Whether a small enough consistency ratio guarantees closure to the true priority vector is also debatable and apart of the convenient algebraic relationship between the components of the priority vector and the elements in the decision matrix, in my opinion should be carefully checked , through large number of experiments, whether improving consistency is pushing closer to the true priority vector. Then the answer to your question will naturally follow.

For this, we calculate the geometric mean of the responses simply by using Ms Excel. 

I used Geometric mean approach. However, if possible, find the ranking of individuals respondents and and compare with the combined ones.

In my experience it is better to compute weights for each participants. Since necessity of inconsistency calculation, consider individuals rather than group responses. So you can calculate inconsistency for each participants and even for any criterion of participants. Finally it is possible to combine all answers to compute total weights. We do these all by EC software.