You could be lazy, and wrap it in a lambda:
my_hash = YAML.load_file('yml')
my_lamb = lambda { |key| my_hash[key.to_s] }
my_lamb[:a] == my_hash['a'] #=> true
But this would only work for reading from the hash - not writing.
To do that, you could use Hash#merge
my_hash = Hash.new { |h,k| h[k] = h[k.to_s] }.merge(YAML.load_file('yml'))
The init block will convert the keys one time on demand, though if you update the value for the string version of the key after accessing the symbol version, the symbol version won't be updated.
irb> x = { 'a' => 1, 'b' => 2 }
#=> {"a"=>1, "b"=>2}
irb> y = Hash.new { |h,k| h[k] = h[k.to_s] }.merge(x)
#=> {"a"=>1, "b"=>2}
irb> y[:a] # the key :a doesn't exist for y, so the init block is called
#=> 1
irb> y
#=> {"a"=>1, :a=>1, "b"=>2}
irb> y[:a] # the key :a now exists for y, so the init block is isn't called
#=> 1
irb> y['a'] = 3
#=> 3
irb> y
#=> {"a"=>3, :a=>1, "b"=>2}
You could also have the init block not update the hash, which would protect you from that kind of error, but you'd still be vulnerable to the opposite - updating the symbol version wouldn't update the string version:
irb> q = { 'c' => 4, 'd' => 5 }
#=> {"c"=>4, "d"=>5}
irb> r = Hash.new { |h,k| h[k.to_s] }.merge(q)
#=> {"c"=>4, "d"=>5}
irb> r[:c] # init block is called
#=> 4
irb> r
#=> {"c"=>4, "d"=>5}
irb> r[:c] # init block is called again, since this key still isn't in r
#=> 4
irb> r[:c] = 7
#=> 7
irb> r
#=> {:c=>7, "c"=>4, "d"=>5}
So the thing to be careful of with these is switching between the two key forms. Stick with one.