The short and sweet answer is "it's accurate in specific ways", but allow me to explain what that means exactly. When we translate any text, a whole bunch of decisions have to be made about how we turn a text that is incomprehensible to our audience into one that they can understand. So, a basic example would be when you have a foreign friend over who doesn't speak your language, and you go to a cafe: you explain your friend's order to the waiter by simply rendering the words they use (e.g. 'thé vert') into their English equivalent ('green tea'). Now, to some extent, this basic principle holds true for literally any translation you ever do. In its most direct form, translation is transposition - changing sounds or written characters from something that's not comprehensible to someone into something that is. Of course, any translator or interpreter worth their salt will tell you that there's a million things that factor in to what 'comprehensible' really means, and that the actual skill in translation lies in understanding what sounds 'right' or 'natural'. To give one example, when you introduce someone in French, the construction you use is technically a reflexive one ('il s'appelle Jean', 'he calls himself Jean'). This literal translation is comprehensible in English, but sounds really clunky. (It also raises further questions, because in English the phrase suggests that Jean isn't actually his real name but just the one he uses... but I digress.)

So even between relatively closely related languages like English and French, which are part of the same language family and influenced each other heavily for centuries, you can run into trouble when all you do is transpose. And this is taking into account that we currently have complete dictionaries, exhaustive grammars, and living speakers of the language who can confirm whether something makes sense in the translation! In taking translation beyond transposition, you inevitably have to stray from the 'raw data', because what you're doing is not just literally finding the equivalent of a French word in English; you're trying to present the translation in such a way that an English person would read it and understand hopefully as many aspects of the text as a French person reading the French text would. Now realistically, this is impossible: there are always going to be aspects of the text - cultural references, word plays, straight up untranslatable words, cadence, rhyme, whatever - that you will miss if you read a translation, so the translator's job is to make decisions about what aspects to stress. A translation of Vergil's Aeneid that preserves the dactylic hexametre in English, for example, might take liberties with the rendition of words to facilitate the rhythm; vice versa, it would be nearly impossible to have a literal translation where all words are translated directly while maintaining metre, but a literal translation approximates the words much better. Both have different functions, of course: if you study poetry and want to study Vergil but don't know any Latin, a translation that prioritises poetic qualities is far more useful than a literal one.

When we look at translations of ancient texts like the Epic of Gilgamesh, this problem is magnified a million times. There's a number of factors that contribute to this, but the most important one is this: Akkadian is a really dead language. It's so dead that its last native speakers had been dead for hundreds of years when the first Latin texts started to be written down. It survived as a scholarly language for a good while (until the first century AD or so), but from the 8th century BC onward, when Aramaic takes over, it's only spoken as a second language. This means that there are lots and lots of questions about how the actual language worked. Because there are no living Akkadian speakers, we can't just ask someone whether a particular translation makes sense, or what a particular word means. In broad lines we understand what's going on, but when it comes to specific words, or particular grammatical features, we're completely in the dark. (It's not quite as bad as some languages, like Hurrian or Sumerian or Ugaritic, where major aspects of the language are a big question mark, but compared to Latin or Greek it's a mess.)

There's another problem that doesn't have to do with the technicalities of the language so much as with genre: we don't really understand Akkadian poetry. Some features are pretty straightforward and are easily identifiable, such as the use of doublets like in Ishtar's Descent (lines 1-3):

To Kurnugi-a, domain of Ereshkigal / did Ishtar, daughter of Sîn, direct her ear; / indeed did the daughter of Sîn direct her ear.

The superfluous nature of the repeated line immediately makes it clear that this is not prose - there's a poetic or lyrical quality to it. It's not always this obvious, though: there are plenty of texts that don't employ doublets (or occasionally have them but not throughout, like Gilgamesh) that still 'feel' like they are poetry, because they use uncommon words, metaphorical language, lots of alliteration or assonance, or other things we might associate with poetry. But what makes a text poetic, and what specific qualities of the text would've been key to an ancient audience for understanding it as such, we don't really know.

So to recap briefly: translation is much more than just transposition because you need to decide on the right balance between literal rendition and conveying the source text as its original audience might've understood it (particularly when the source text is poetry). When it comes to ancient literature, both of these are more difficult because there are so many aspects of the source text that we don't understand.

---

Now how does this help us understand Mitchell's translation of the Epic of Gilgamesh? Well, when I called it "accurate in specific ways", I meant that it's a translation that tries to tell us the story with the same impact as he thinks the ancient audience would've received it. It's not a word-for-word translation - and to compare have a look at Andrew George, The Babylonian Gilgamesh Epic (2003), which is scholarly and tremendously dry - but rather a way of making a modern audience appreciate the story and its poetic qualities as if it were a modern composition rather than an ancient one. It's accurate in that it's a tremendously readable version of the epic that conveys it to a broad audience better than any literal translation could, regardless of how true that translation might be to the words of the text.

As a final note: I understand that Mitchell synthesised English translations and didn't actually translate it from Akkadian himself - this is absolutely fine, just another way of bringing a particular story to a broad audience.