There are many mundane reasons you can come up with (e.g. cost) - and there's no harm in including them as additional supportive factors. However, the main reason I personally favour is a rather philosophical one: using humans to fight wars directly is considered to be the best way to discourage war and large scale bloodshed.
This may seem a little counter-intuitive, so I'll give you a real-world analogy.
If you were around (and politically aware) during the Reagan era, you'll have heard of the Strategic Defense Initiative (colloquially derided as "Star Wars"). This was basically a laser-based ballistic missile defence system. The ultimate goal was to have a system so good that it could intercept every inbound nuclear-tipped ICBM (intercontinental ballistic missile) before it could detonate over US soil.
Putting aside logistics and efficacy, mooting such a system seems like a really obvious thing. Given the immense destructive power of strategic nuclear weapons, how can one ever argue against an effective defence against them?
But one of the main arguments against the SDI was that it would destabilise the delicate balance of power between the US and the USSR (the only serious nuclear rival at the time). Because the nuclear assets were considered so evenly matched, the powers were considered to be held in check by the shared concern of Mutually Assured Destruction (or MAD for short, a very appropriate acronym). The main thing holding one country from starting a nuke strike on the other was the certainty of being utterly destroyed in response. But if one country were to develop an effective physical defence against an incoming nuclear strike, then that country would immediately have the advantage and could (perhaps) contemplate a first strike with relative impunity. And (thinking further), even allowing one country to start making serious inroads into such a programme might prompt its chief enemy to jump the gun and try its lucky by launching a pre-emptive strike while things were still relatively "even". That's actually the sort of public political and military debate that was going on at the time.
In the end, the SDI never took off, so all that's academic. But, as mad as it sounds, MAD remains an accepted doctrine.
Now let's see how that can be adapted to your scenario. War should be seen as the last alternative given the lasting damage it does to human societies. Let's say your fictional world is comprised of democratic states (not psychopathic despotic regimes) in a state of delicate military balance with existing military forces more or less serving as an effective enough deterrent against attack by others. However, tensions are always simmering and countries would be quite happy to seize foreign territory and resources if it were feasible to do so.
A country that develops super-efficient and practically invulnerable "killerbots" before any other can immediately start attacking others and annexing territory with relative impunity. In fact, several countries have realised this and started up independent killerbot initiatives, all racing to get the first working prototype robo-army up and running (and killing).
The World Council (like the UN in our world, only more effective) sees this as a very worrying trend. It decrees that the manufacture of killerbots is absolutely outlawed by international convention. Only human soldiers are allowed (and only for defensive purposes). Their reasoning goes along the same lines as the objections to the SDI in our world: human life is considered universally precious (remember - no despotic tyrannies in your world) and the thought of losing their able-bodied young people en masse is repulsive to every nation. That becomes the biggest deterrent to wars of aggression - analogous to MAD. However, allowing killerbot programmes to progress undermines this deterrent. Therefore, the killerbot programmes must be outlawed. And analogous to the second-order thinking related to the SDI that I mentioned, the council also opined that if one country was seen as having leapfrogged over others in coming close to a functioning killerbot, others might immediately attack pre-emptively just to safeguard themselves as they would consider the immediate limited price in blood to be more judicious than a much bigger one down the road (at the white-hot metallic pincers of the killerbots).
So killerbots have been universally outlawed by treaty (you can think of a way of effectively surveilling for treaty-breakers), leaving only human soldiers to fight wars (purely defensive ones, as per the intent of the treaty).