我想你需要的是 jsonb_to_recordset 和横向连接。
对于以下架构和数据

CREATE TABLE profiles (
  id integer,
  item_type text,
  object_changes jsonb
);

INSERT INTO profiles2(id, item_type, object_changes) VALUES
  (1, 'Item', '[{"customer_id": [1, 5], "other_id": 1}, {"customer_id": [4, 5], "other_id": 2}]'::jsonb),
  (2, 'Item', '[{"customer_id": [3, 6], "other_id": 3}, {"customer_id": [3, 5], "other_id": 2}]'::jsonb),
  (3, 'Item', '[{"customer_id": [4, 7], "other_id": 3}, {"customer_id": [8, 9], "other_id": 2}]'::jsonb);

类似这样的方法会奏效：

SELECT distinct profiles.*
FROM 
  profiles, 
  jsonb_to_recordset(profiles.object_changes) AS changes(customer_id integer[], other_id integer)
WHERE 5 = ANY(changes.customer_id);

 id | item_type |                                  object_changes
----+-----------+----------------------------------------------------------------------------------
  2 | Item      | [{"other_id": 3, "customer_id": [3, 6]}, {"other_id": 2, "customer_id": [3, 5]}]
  1 | Item      | [{"other_id": 1, "customer_id": [1, 5]}, {"other_id": 2, "customer_id": [4, 5]}]
(2 rows)

因此，使用ar查询接口的最终解决方案类似于（我对要查找的值进行了硬编码，但我相信您已经明白了这一点，参数化不是问题）：

Profile.find_by_sql(<<~SQL)
  SELECT distinct profiles.*
  FROM 
    profiles, 
    jsonb_to_recordset(profiles.object_changes) AS changes(customer_id integer[], other_id integer)
  WHERE 5 = ANY(changes.customer_id)
SQL